{"schemaVersion":3,"generatedAt":"2026-07-27T10:09:31.240281+00:00","pricing":{"schema_version":1,"captured_at":"2026-07-24T03:17:33Z","product":"RunPod GPU Pods","billing_unit":"USD per GPU-hour","default_cloud_tier":"secure","api_source":"RunPod GPU type API","public_source":"https://www.runpod.io/pricing","public_source_updated_at":"2026-07-17","notes":["Rates are on-demand Pod compute prices and can change.","Storage, setup time, idle time, data preparation, checkpoint I/O, and taxes are excluded.","Community and Secure Cloud prices are retained separately.","The benchmark is single-GPU; prices do not imply multi-GPU scaling efficiency."],"gpus":{"NVIDIA GeForce RTX 4090":{"display_name":"RTX 4090","vram_gb":24,"community_usd_per_hour":0.34,"secure_usd_per_hour":0.69},"NVIDIA GeForce RTX 5090":{"display_name":"RTX 5090","vram_gb":32,"community_usd_per_hour":0.69,"secure_usd_per_hour":0.99},"NVIDIA RTX A6000":{"display_name":"RTX A6000","vram_gb":48,"community_usd_per_hour":0.33,"secure_usd_per_hour":0.53},"NVIDIA RTX 6000 Ada Generation":{"display_name":"RTX 6000 Ada","vram_gb":48,"community_usd_per_hour":0.74,"secure_usd_per_hour":0.84},"NVIDIA RTX PRO 6000 Blackwell Server Edition":{"display_name":"RTX PRO 6000","vram_gb":96,"community_usd_per_hour":1.69,"secure_usd_per_hour":1.99},"NVIDIA A100 80GB PCIe":{"display_name":"A100 PCIe 80GB","vram_gb":80,"community_usd_per_hour":1.19,"secure_usd_per_hour":1.39},"NVIDIA H100 80GB HBM3":{"display_name":"H100 SXM 80GB","vram_gb":80,"community_usd_per_hour":2.69,"secure_usd_per_hour":2.99},"NVIDIA H200":{"display_name":"H200 SXM 141GB","vram_gb":141,"community_usd_per_hour":3.59,"secure_usd_per_hour":4.39},"NVIDIA B200":{"display_name":"B200 180GB","vram_gb":180,"community_usd_per_hour":5.98,"secure_usd_per_hour":5.89},"NVIDIA B300 SXM6 AC":{"display_name":"B300 288GB","vram_gb":288,"community_usd_per_hour":6.94,"secure_usd_per_hour":7.39}}},"providerPricing":{"captured_at":"2026-07-27T05:22:34+00:00","comparison_policy":{"benchmark_transfer":"Prices outside RunPod are projections using measured RunPod TPS, not measurements on the quoted provider.","excluded_costs":["persistent storage unless inseparable from the instance","network egress","tax","setup and idle time","commitment discounts"],"included_costs":"advertised accelerator or bundled-instance rate","vast_quote_rule":"Median of the three cheapest verified, rentable, single-GPU, on-demand offers with reliability >=0.98, CUDA >=12.8, and VRAM matching the measured SKU.","warning":"Provider products differ in CPU, RAM, storage, networking, availability, tenancy, and minimum rental commitment."},"currency":"USD","quotes":[{"gpu":"NVIDIA A100 80GB PCIe","instance_gpu_count":8,"match_quality":"near","notes":"A100 80 GB; instance fabric/clock may differ from measured PCIe host.","price_type":"capacity_block","product":"p4de.24xlarge","provider":"AWS","region":"us-east-1","source_url":"https://aws.amazon.com/ec2/capacityblocks/pricing/","usd_per_gpu_hour":2.214},{"gpu":"NVIDIA A100 80GB PCIe","instance_gpu_count":1,"match_quality":"near","notes":"Full A100 80 GB VM; interconnect/clock may differ from measured PCIe host.","price_type":"on_demand","product":"a2-ultragpu-1g","provider":"Google Cloud","region":"us-central1","source_url":"https://cloud.google.com/products/compute/pricing/accelerator-optimized","usd_per_gpu_hour":5.06879789},{"gpu":"NVIDIA A100 80GB PCIe","instance_gpu_count":8,"match_quality":"near","notes":"Measured device is PCIe; Lambda quote is an 8-GPU SXM instance.","price_type":"on_demand","product":"8x A100 SXM 80 GB","provider":"Lambda","region":null,"source_url":"https://lambda.ai/instances","usd_per_gpu_hour":2.79},{"gpu":"NVIDIA A100 80GB PCIe","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"marketplace_on_demand","product":"A100 PCIe 80GB","provider":"RunPod Community","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":1.19},{"gpu":"NVIDIA A100 80GB PCIe","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"A100 PCIe 80GB","provider":"RunPod Secure","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":1.39},{"gpu":"NVIDIA A100 80GB PCIe","instance_gpu_count":1,"match_quality":"exact","notes":"Median of 3 cheapest eligible offers; range $0.867--$1.000/GPU-hour.","price_type":"marketplace_on_demand","product":"A100 PCIE","provider":"Vast.ai","region":null,"source_url":"https://docs.vast.ai/api-reference/search/search-offers","usd_per_gpu_hour":0.988},{"gpu":"NVIDIA B200","instance_gpu_count":8,"match_quality":"exact","notes":"","price_type":"capacity_block","product":"p6-b200.48xlarge","provider":"AWS","region":"us-east-1","source_url":"https://aws.amazon.com/ec2/capacityblocks/pricing/","usd_per_gpu_hour":12.355},{"gpu":"NVIDIA B200","instance_gpu_count":8,"match_quality":"exact","notes":"No ordinary on-demand price was listed at capture time.","price_type":"flex_start","product":"a4-highgpu-8g","provider":"Google Cloud","region":"us-central1","source_url":"https://cloud.google.com/products/compute/pricing/accelerator-optimized","usd_per_gpu_hour":8.055},{"gpu":"NVIDIA B200","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"1x B200 SXM6 180 GB","provider":"Lambda","region":null,"source_url":"https://lambda.ai/instances","usd_per_gpu_hour":6.99},{"gpu":"NVIDIA B200","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"marketplace_on_demand","product":"B200 180GB","provider":"RunPod Community","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":5.98},{"gpu":"NVIDIA B200","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"B200 180GB","provider":"RunPod Secure","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":5.89},{"gpu":"NVIDIA B200","instance_gpu_count":1,"match_quality":"exact","notes":"Median of 3 cheapest eligible offers; range $6.237--$6.487/GPU-hour.","price_type":"marketplace_on_demand","product":"B200","provider":"Vast.ai","region":null,"source_url":"https://docs.vast.ai/api-reference/search/search-offers","usd_per_gpu_hour":6.25},{"gpu":"NVIDIA B300 SXM6 AC","instance_gpu_count":8,"match_quality":"exact","notes":"","price_type":"capacity_block","product":"p6-b300.48xlarge","provider":"AWS","region":"us-east-1","source_url":"https://aws.amazon.com/ec2/capacityblocks/pricing/","usd_per_gpu_hour":14.04},{"gpu":"NVIDIA B300 SXM6 AC","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"marketplace_on_demand","product":"B300 288GB","provider":"RunPod Community","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":6.94},{"gpu":"NVIDIA B300 SXM6 AC","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"B300 288GB","provider":"RunPod Secure","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":7.39},{"gpu":"NVIDIA GeForce RTX 4090","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"marketplace_on_demand","product":"RTX 4090","provider":"RunPod Community","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":0.34},{"gpu":"NVIDIA GeForce RTX 4090","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"RTX 4090","provider":"RunPod Secure","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":0.69},{"gpu":"NVIDIA GeForce RTX 4090","instance_gpu_count":1,"match_quality":"exact","notes":"Median of 3 cheapest eligible offers; range $0.253--$0.253/GPU-hour.","price_type":"marketplace_on_demand","product":"RTX 4090","provider":"Vast.ai","region":null,"source_url":"https://docs.vast.ai/api-reference/search/search-offers","usd_per_gpu_hour":0.2533333333333333},{"gpu":"NVIDIA GeForce RTX 5090","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"marketplace_on_demand","product":"RTX 5090","provider":"RunPod Community","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":0.69},{"gpu":"NVIDIA GeForce RTX 5090","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"RTX 5090","provider":"RunPod Secure","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":0.99},{"gpu":"NVIDIA GeForce RTX 5090","instance_gpu_count":1,"match_quality":"exact","notes":"Median of 3 cheapest eligible offers; range $0.280--$0.307/GPU-hour.","price_type":"marketplace_on_demand","product":"RTX 5090","provider":"Vast.ai","region":null,"source_url":"https://docs.vast.ai/api-reference/search/search-offers","usd_per_gpu_hour":0.29333333333333333},{"gpu":"NVIDIA H100 80GB HBM3","instance_gpu_count":1,"match_quality":"exact","notes":"Single-H100 Capacity Block effective hourly rate.","price_type":"capacity_block","product":"p5.4xlarge","provider":"AWS","region":"us-east-1","source_url":"https://aws.amazon.com/ec2/capacityblocks/pricing/","usd_per_gpu_hour":5.191},{"gpu":"NVIDIA H100 80GB HBM3","instance_gpu_count":8,"match_quality":"exact","notes":"","price_type":"on_demand","product":"a3-highgpu-8g","provider":"Google Cloud","region":"us-central1","source_url":"https://cloud.google.com/products/compute/pricing/accelerator-optimized","usd_per_gpu_hour":11.061250014875},{"gpu":"NVIDIA H100 80GB HBM3","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"1x H100 SXM 80 GB","provider":"Lambda","region":null,"source_url":"https://lambda.ai/instances","usd_per_gpu_hour":4.29},{"gpu":"NVIDIA H100 80GB HBM3","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"marketplace_on_demand","product":"H100 SXM 80GB","provider":"RunPod Community","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":2.69},{"gpu":"NVIDIA H100 80GB HBM3","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"H100 SXM 80GB","provider":"RunPod Secure","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":2.99},{"gpu":"NVIDIA H100 80GB HBM3","instance_gpu_count":1,"match_quality":"exact","notes":"Median of 3 cheapest eligible offers; range $2.267--$3.000/GPU-hour.","price_type":"marketplace_on_demand","product":"H100 SXM","provider":"Vast.ai","region":null,"source_url":"https://docs.vast.ai/api-reference/search/search-offers","usd_per_gpu_hour":2.5453333333333332},{"gpu":"NVIDIA H200","instance_gpu_count":8,"match_quality":"exact","notes":"","price_type":"capacity_block","product":"p5e.48xlarge","provider":"AWS","region":"us-east-2","source_url":"https://aws.amazon.com/ec2/capacityblocks/pricing/","usd_per_gpu_hour":5.97},{"gpu":"NVIDIA H200","instance_gpu_count":8,"match_quality":"exact","notes":"","price_type":"on_demand","product":"a3-ultragpu-8g","provider":"Google Cloud","region":"us-central1","source_url":"https://cloud.google.com/products/compute/pricing/accelerator-optimized","usd_per_gpu_hour":10.600863561625},{"gpu":"NVIDIA H200","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"marketplace_on_demand","product":"H200 SXM 141GB","provider":"RunPod Community","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":3.59},{"gpu":"NVIDIA H200","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"H200 SXM 141GB","provider":"RunPod Secure","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":4.39},{"gpu":"NVIDIA H200","instance_gpu_count":1,"match_quality":"exact","notes":"Median of 3 cheapest eligible offers; range $3.934--$4.287/GPU-hour.","price_type":"marketplace_on_demand","product":"H200","provider":"Vast.ai","region":null,"source_url":"https://docs.vast.ai/api-reference/search/search-offers","usd_per_gpu_hour":4.078947368421053},{"gpu":"NVIDIA RTX 6000 Ada Generation","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"marketplace_on_demand","product":"RTX 6000 Ada","provider":"RunPod Community","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":0.74},{"gpu":"NVIDIA RTX 6000 Ada Generation","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"RTX 6000 Ada","provider":"RunPod Secure","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":0.84},{"gpu":"NVIDIA RTX 6000 Ada Generation","instance_gpu_count":1,"match_quality":"exact","notes":"Median of 3 cheapest eligible offers; range $0.533--$0.560/GPU-hour.","price_type":"marketplace_on_demand","product":"RTX 6000Ada","provider":"Vast.ai","region":null,"source_url":"https://docs.vast.ai/api-reference/search/search-offers","usd_per_gpu_hour":0.5479999999999999},{"gpu":"NVIDIA RTX A6000","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"1x RTX A6000","provider":"Lambda","region":null,"source_url":"https://lambda.ai/instances","usd_per_gpu_hour":1.09},{"gpu":"NVIDIA RTX A6000","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"marketplace_on_demand","product":"RTX A6000","provider":"RunPod Community","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":0.33},{"gpu":"NVIDIA RTX A6000","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"RTX A6000","provider":"RunPod Secure","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":0.53},{"gpu":"NVIDIA RTX A6000","instance_gpu_count":1,"match_quality":"exact","notes":"Median of 1 cheapest eligible offers; range $0.387--$0.387/GPU-hour.","price_type":"marketplace_on_demand","product":"RTX A6000","provider":"Vast.ai","region":null,"source_url":"https://docs.vast.ai/api-reference/search/search-offers","usd_per_gpu_hour":0.38666666666666666},{"gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"g4-standard-48","provider":"Google Cloud","region":"us-central1","source_url":"https://cloud.google.com/products/compute/pricing/accelerator-optimized","usd_per_gpu_hour":4.49993},{"gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"marketplace_on_demand","product":"RTX PRO 6000","provider":"RunPod Community","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":1.69},{"gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","instance_gpu_count":1,"match_quality":"exact","notes":"","price_type":"on_demand","product":"RTX PRO 6000","provider":"RunPod Secure","region":null,"source_url":"https://www.runpod.io/pricing","usd_per_gpu_hour":1.99},{"gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","instance_gpu_count":1,"match_quality":"exact","notes":"Median of 3 cheapest eligible offers; range $1.267--$1.333/GPU-hour.","price_type":"marketplace_on_demand","product":"RTX PRO 6000 S","provider":"Vast.ai","region":null,"source_url":"https://docs.vast.ai/api-reference/search/search-offers","usd_per_gpu_hour":1.2666666666666666}],"schema_version":1,"sources":["https://aws.amazon.com/ec2/capacityblocks/pricing/","https://cloud.google.com/products/compute/pricing/accelerator-optimized","https://docs.vast.ai/api-reference/search/search-offers","https://lambda.ai/instances","https://www.runpod.io/pricing"],"unit":"per GPU-hour","vast_query_audit":{"NVIDIA A100 80GB PCIe":{"cheapest_three_sanitized":[{"country_code":"US","cuda_max_good":13.0,"reliability":0.9981,"total_usd_per_hour_at_listed_storage":0.8733333333333334,"usd_per_gpu_hour":0.8666666666666667,"vram_mb":81920.0},{"country_code":"JP","cuda_max_good":13.3,"reliability":0.992683,"total_usd_per_hour_at_listed_storage":0.9914311111111112,"usd_per_gpu_hour":0.988,"vram_mb":81920.0},{"country_code":"GB","cuda_max_good":13.1,"reliability":0.9896447,"total_usd_per_hour_at_listed_storage":1.0022222222222221,"usd_per_gpu_hour":1.0,"vram_mb":81920.0}],"eligible_offer_count":6,"query_gpu_name":"A100 PCIE","quote_rule":"median of three cheapest eligible offers"},"NVIDIA B200":{"cheapest_three_sanitized":[{"country_code":"US","cuda_max_good":13.2,"reliability":0.9976631,"total_usd_per_hour_at_listed_storage":6.247916666666667,"usd_per_gpu_hour":6.2375,"vram_mb":183359.0},{"country_code":"US","cuda_max_good":13.0,"reliability":0.9981403,"total_usd_per_hour_at_listed_storage":6.252083333333333,"usd_per_gpu_hour":6.25,"vram_mb":183359.0},{"country_code":"US","cuda_max_good":13.2,"reliability":0.9957008,"total_usd_per_hour_at_listed_storage":6.497916666666667,"usd_per_gpu_hour":6.4875,"vram_mb":183359.0}],"eligible_offer_count":5,"query_gpu_name":"B200","quote_rule":"median of three cheapest eligible offers"},"NVIDIA B300 SXM6 AC":{"cheapest_three_sanitized":[],"eligible_offer_count":0,"query_gpu_name":"B300","quote_rule":"median of three cheapest eligible offers"},"NVIDIA GeForce RTX 4090":{"cheapest_three_sanitized":[{"country_code":"US","cuda_max_good":13.0,"reliability":0.9955969,"total_usd_per_hour_at_listed_storage":0.26296296296296295,"usd_per_gpu_hour":0.2533333333333333,"vram_mb":24564.0},{"country_code":"US","cuda_max_good":13.0,"reliability":0.9961348,"total_usd_per_hour_at_listed_storage":0.26296296296296295,"usd_per_gpu_hour":0.2533333333333333,"vram_mb":24564.0},{"country_code":"US","cuda_max_good":13.0,"reliability":0.9933616,"total_usd_per_hour_at_listed_storage":0.26296296296296295,"usd_per_gpu_hour":0.2533333333333333,"vram_mb":24564.0}],"eligible_offer_count":38,"query_gpu_name":"RTX 4090","quote_rule":"median of three cheapest eligible offers"},"NVIDIA GeForce RTX 5090":{"cheapest_three_sanitized":[{"country_code":"US","cuda_max_good":13.0,"reliability":0.9933666,"total_usd_per_hour_at_listed_storage":0.2896296296296297,"usd_per_gpu_hour":0.28,"vram_mb":32607.0},{"country_code":"US","cuda_max_good":13.0,"reliability":0.9945708,"total_usd_per_hour_at_listed_storage":0.302962962962963,"usd_per_gpu_hour":0.29333333333333333,"vram_mb":32607.0},{"country_code":"US","cuda_max_good":13.0,"reliability":0.9949772,"total_usd_per_hour_at_listed_storage":0.30814814814814817,"usd_per_gpu_hour":0.3066666666666667,"vram_mb":32607.0}],"eligible_offer_count":55,"query_gpu_name":"RTX 5090","quote_rule":"median of three cheapest eligible offers"},"NVIDIA H100 80GB HBM3":{"cheapest_three_sanitized":[{"country_code":"MY","cuda_max_good":13.2,"reliability":0.9905361,"total_usd_per_hour_at_listed_storage":2.2696296296296294,"usd_per_gpu_hour":2.2666666666666666,"vram_mb":81559.0},{"country_code":"CZ","cuda_max_good":13.2,"reliability":0.9958913,"total_usd_per_hour_at_listed_storage":2.551851851851852,"usd_per_gpu_hour":2.5453333333333332,"vram_mb":81559.0},{"country_code":"TW","cuda_max_good":13.0,"reliability":0.9834368,"total_usd_per_hour_at_listed_storage":3.005333333333333,"usd_per_gpu_hour":3.0,"vram_mb":81559.0}],"eligible_offer_count":8,"query_gpu_name":"H100 SXM","quote_rule":"median of three cheapest eligible offers"},"NVIDIA H200":{"cheapest_three_sanitized":[{"country_code":"SA","cuda_max_good":13.0,"reliability":0.9974872,"total_usd_per_hour_at_listed_storage":3.9356725146198834,"usd_per_gpu_hour":3.93421052631579,"vram_mb":143771.0},{"country_code":"US","cuda_max_good":12.8,"reliability":0.9994338,"total_usd_per_hour_at_listed_storage":4.0811403508771935,"usd_per_gpu_hour":4.078947368421053,"vram_mb":143771.0},{"country_code":"JP","cuda_max_good":13.2,"reliability":0.9971439,"total_usd_per_hour_at_listed_storage":4.2911900584795335,"usd_per_gpu_hour":4.286842105263159,"vram_mb":143771.0}],"eligible_offer_count":3,"query_gpu_name":"H200","quote_rule":"median of three cheapest eligible offers"},"NVIDIA RTX 6000 Ada Generation":{"cheapest_three_sanitized":[{"country_code":"CN","cuda_max_good":13.0,"reliability":0.9961997,"total_usd_per_hour_at_listed_storage":0.5355555555555556,"usd_per_gpu_hour":0.5333333333333333,"vram_mb":49140.0},{"country_code":"JP","cuda_max_good":13.0,"reliability":0.9939217,"total_usd_per_hour_at_listed_storage":0.5502222222222222,"usd_per_gpu_hour":0.5479999999999999,"vram_mb":49140.0},{"country_code":"US","cuda_max_good":13.0,"reliability":0.9968621,"total_usd_per_hour_at_listed_storage":0.5659259259259259,"usd_per_gpu_hour":0.56,"vram_mb":49140.0}],"eligible_offer_count":5,"query_gpu_name":"RTX 6000Ada","quote_rule":"median of three cheapest eligible offers"},"NVIDIA RTX A6000":{"cheapest_three_sanitized":[{"country_code":"RO","cuda_max_good":13.0,"reliability":0.9916833,"total_usd_per_hour_at_listed_storage":0.3888888888888889,"usd_per_gpu_hour":0.38666666666666666,"vram_mb":49140.0}],"eligible_offer_count":1,"query_gpu_name":"RTX A6000","quote_rule":"median of three cheapest eligible offers"},"NVIDIA RTX PRO 6000 Blackwell Server Edition":{"cheapest_three_sanitized":[{"country_code":"AU","cuda_max_good":13.2,"reliability":0.9982606,"total_usd_per_hour_at_listed_storage":1.2688888888888887,"usd_per_gpu_hour":1.2666666666666666,"vram_mb":97887.0},{"country_code":"US","cuda_max_good":13.2,"reliability":0.9967968,"total_usd_per_hour_at_listed_storage":1.2688888888888887,"usd_per_gpu_hour":1.2666666666666666,"vram_mb":97887.0},{"country_code":"BG","cuda_max_good":13.0,"reliability":0.9978267,"total_usd_per_hour_at_listed_storage":1.3362962962962963,"usd_per_gpu_hour":1.3333333333333333,"vram_mb":97887.0}],"eligible_offer_count":10,"query_gpu_name":"RTX PRO 6000 S","quote_rule":"median of three cheapest eligible offers"}}},"provenance":{"catalog":"data/benchmark_catalog.csv","catalogRows":1342,"supplementalRows":1621,"supplementalSources":["artifacts/runpod-snapshots/20260723_rtx6000_ada_48gb_paused/olm-mfu-results/rtx6000_ada_context_frontier_seed11_v2.json","artifacts/runpod-snapshots/20260723_rtx6000_ada_48gb_paused/olm-mfu-results/rtx6000_ada_training_seed11_v2.json","artifacts/runpod-snapshots/20260723_rtx6000_ada_48gb_paused/olm-mfu-results/rtx6000_ada_training_seed22_v2.json","artifacts/runpod-snapshots/20260724_a6000_48gb/olm-mfu-a6000/results/a6000_context_extension_16384_seed11_v2.json","artifacts/runpod-snapshots/20260724_a6000_48gb/olm-mfu-a6000/results/a6000_context_extension_16384_seed22_v2.json","artifacts/runpod-snapshots/20260724_a6000_48gb/olm-mfu-a6000/results/a6000_context_extension_32768_seed11_v2.json","artifacts/runpod-snapshots/20260724_a6000_48gb/olm-mfu-a6000/results/a6000_context_extension_32768_seed22_v2.json","artifacts/runpod-snapshots/20260724_a6000_48gb/olm-mfu-a6000/results/a6000_context_frontier_seed11_v2.json","artifacts/runpod-snapshots/20260724_a6000_48gb/olm-mfu-a6000/results/a6000_context_frontier_seed22_v2.json","artifacts/runpod-snapshots/20260724_a6000_48gb/olm-mfu-a6000/results/a6000_training_seed11_v2.json","artifacts/runpod-snapshots/20260724_a6000_48gb/olm-mfu-a6000/results/a6000_training_seed22_v2.json","artifacts/runpod-snapshots/20260724_b200_180gb/olm-mfu-b200/results/b200_context_extension_16384_seed11_v2.json","artifacts/runpod-snapshots/20260724_b200_180gb/olm-mfu-b200/results/b200_context_extension_16384_seed22_v2.json","artifacts/runpod-snapshots/20260724_b200_180gb/olm-mfu-b200/results/b200_context_extension_32768_seed11_v2.json","artifacts/runpod-snapshots/20260724_b200_180gb/olm-mfu-b200/results/b200_context_extension_32768_seed22_v2.json","artifacts/runpod-snapshots/20260724_b200_180gb/olm-mfu-b200/results/b200_context_frontier_seed11_v2.json","artifacts/runpod-snapshots/20260724_b200_180gb/olm-mfu-b200/results/b200_context_frontier_seed22_v2.json","artifacts/runpod-snapshots/20260724_b200_180gb/olm-mfu-b200/results/b200_training_seed11_v2.json","artifacts/runpod-snapshots/20260724_b200_180gb/olm-mfu-b200/results/b200_training_seed22_v2.json","artifacts/runpod-snapshots/20260724_b300_sxm6_ac_288gb_budget_scoped/olm-mfu-b300/results/b300_context_frontier_seed11_v2.json","artifacts/runpod-snapshots/20260724_b300_sxm6_ac_288gb_budget_scoped/olm-mfu-b300/results/b300_context_frontier_seed22_v2.json","artifacts/runpod-snapshots/20260724_b300_sxm6_ac_288gb_budget_scoped/olm-mfu-b300/results/b300_training_seed11_v2.json","artifacts/runpod-snapshots/20260724_b300_sxm6_ac_288gb_budget_scoped/olm-mfu-b300/results/b300_training_seed22_v2.json","artifacts/runpod-snapshots/20260724_b300_sxm6_ac_288gb_complete/olm-mfu-b300/results/b300_context_extension_16384_seed11_v2.json","artifacts/runpod-snapshots/20260724_b300_sxm6_ac_288gb_complete/olm-mfu-b300/results/b300_context_extension_16384_seed22_v2.json","artifacts/runpod-snapshots/20260724_b300_sxm6_ac_288gb_complete/olm-mfu-b300/results/b300_context_extension_32768_seed11_v2.json","artifacts/runpod-snapshots/20260724_b300_sxm6_ac_288gb_complete/olm-mfu-b300/results/b300_context_extension_32768_seed22_v2.json","artifacts/runpod-snapshots/20260724_h200_sxm_141gb/olm-mfu-h200/results/h200_context_extension_16384_seed11_v2.json","artifacts/runpod-snapshots/20260724_h200_sxm_141gb/olm-mfu-h200/results/h200_context_extension_16384_seed22_v2.json","artifacts/runpod-snapshots/20260724_h200_sxm_141gb/olm-mfu-h200/results/h200_context_extension_32768_seed11_v2.json","artifacts/runpod-snapshots/20260724_h200_sxm_141gb/olm-mfu-h200/results/h200_context_extension_32768_seed22_v2.json","artifacts/runpod-snapshots/20260724_h200_sxm_141gb/olm-mfu-h200/results/h200_context_frontier_seed11_v2.json","artifacts/runpod-snapshots/20260724_h200_sxm_141gb/olm-mfu-h200/results/h200_context_frontier_seed22_v2.json","artifacts/runpod-snapshots/20260724_h200_sxm_141gb/olm-mfu-h200/results/h200_training_seed11_v2.json","artifacts/runpod-snapshots/20260724_h200_sxm_141gb/olm-mfu-h200/results/h200_training_seed22_v2.json"],"totalRows":2963,"completeRows":2105,"method":"Canonical catalog plus deduplicated completed publication-protocol JSON from local RunPod safety snapshots."},"protocol":{"trainingStep":"BF16-autocast forward, shifted-token cross-entropy, backward, and fused AdamW update","parameterStorage":"FP32","gradientStorage":"FP32","optimizerStateStorage":"FP32","activationCheckpointing":false,"gradientAccumulationSteps":1,"measuredSteps":20,"seeds":[11,22],"mfuFormula":"6N + 12LHS FLOPs per token"},"models":[{"key":"150m","label":"150M","parameters":150436608},{"key":"250m","label":"250M","parameters":246397312},{"key":"350m","label":"350M","parameters":348447744},{"key":"500m","label":"500M","parameters":505516800},{"key":"700m","label":"700M","parameters":707480064},{"key":"1b","label":"1B","parameters":995135232},{"key":"2b","label":"2B","parameters":2008824320},{"key":"4b","label":"4B","parameters":3999611392},{"key":"8b","label":"8B","parameters":7983108096}],"contexts":[512,1024,2048,4096,8192,16384,32768],"gpus":[{"id":"NVIDIA RTX A6000","short_name":"RTX A6000","family":"Ampere","color":"#225c73","vramGb":48,"secureUsdPerHour":0.53,"communityUsdPerHour":0.33,"seeds":[11,22],"completeRows":164,"totalRows":258},{"id":"NVIDIA GeForce RTX 4090","short_name":"RTX 4090","family":"Ada","color":"#b45b2b","vramGb":24,"secureUsdPerHour":0.69,"communityUsdPerHour":0.34,"seeds":[11,22],"completeRows":110,"totalRows":176},{"id":"NVIDIA RTX 6000 Ada Generation","short_name":"RTX 6000 Ada","family":"Ada","color":"#476b42","vramGb":48,"secureUsdPerHour":0.84,"communityUsdPerHour":0.74,"seeds":[11,22],"completeRows":113,"totalRows":137},{"id":"NVIDIA GeForce RTX 5090","short_name":"RTX 5090","family":"Blackwell","color":"#b30000","vramGb":32,"secureUsdPerHour":0.99,"communityUsdPerHour":0.69,"seeds":[11,22],"completeRows":132,"totalRows":210},{"id":"NVIDIA A100 80GB PCIe","short_name":"A100 PCIe","family":"Ampere","color":"#366a9f","vramGb":80,"secureUsdPerHour":1.39,"communityUsdPerHour":1.19,"seeds":[11,22],"completeRows":228,"totalRows":324},{"id":"NVIDIA RTX PRO 6000 Blackwell Server Edition","short_name":"RTX PRO 6000","family":"Blackwell","color":"#6b4c7a","vramGb":96,"secureUsdPerHour":1.99,"communityUsdPerHour":1.69,"seeds":[11,22],"completeRows":266,"totalRows":308},{"id":"NVIDIA H100 80GB HBM3","short_name":"H100 SXM","family":"Hopper","color":"#8a6d00","vramGb":80,"secureUsdPerHour":2.99,"communityUsdPerHour":2.69,"seeds":[11,22],"completeRows":228,"totalRows":324},{"id":"NVIDIA H200","short_name":"H200 SXM","family":"Hopper","color":"#9c4d20","vramGb":141,"secureUsdPerHour":4.39,"communityUsdPerHour":3.59,"seeds":[11,22],"completeRows":260,"totalRows":380},{"id":"NVIDIA B200","short_name":"B200","family":"Blackwell","color":"#00727a","vramGb":180,"secureUsdPerHour":5.89,"communityUsdPerHour":5.98,"seeds":[11,22],"completeRows":292,"totalRows":414},{"id":"NVIDIA B300 SXM6 AC","short_name":"B300","family":"Blackwell","color":"#5f7427","vramGb":288,"secureUsdPerHour":7.39,"communityUsdPerHour":6.94,"seeds":[11,22],"completeRows":312,"totalRows":432}],"rows":[{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":71708.8021446755,"meanTps":71639.91966980495,"stepMs":228.47962188720703,"jitter":0.0037870378616342497,"achievedTflops":71.49306535291763,"nominalPeakTflops":312.0,"mfuNominalPct":22.914444023371036,"configuredPeakTflops":312.0,"mfuConfiguredPct":22.914444023371036,"vramAllocatedGb":22.055896576,"vramAllocatedPct":25.955979885724556,"vramReservedGb":23.200792576,"vramReservedPct":27.303324685100463,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":74701.26737448179,"meanTps":74679.45631036143,"stepMs":438.65386962890625,"jitter":0.0034987615183448228,"achievedTflops":74.47652771517045,"nominalPeakTflops":312.0,"mfuNominalPct":23.870681959990527,"configuredPeakTflops":312.0,"mfuConfiguredPct":23.870681959990527,"vramAllocatedGb":41.987277824,"vramAllocatedPct":49.411772262386975,"vramReservedGb":44.369444864,"vramReservedPct":52.21517132447532,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":54354.347892445054,"meanTps":53489.44902309589,"stepMs":1205.7177124023438,"jitter":0.04465404266040122,"achievedTflops":54.190822184558456,"nominalPeakTflops":312.0,"mfuNominalPct":17.368853264281558,"configuredPeakTflops":312.0,"mfuConfiguredPct":17.368853264281558,"vramAllocatedGb":81.850040832,"vramAllocatedPct":96.32335761824736,"vramReservedGb":82.269175808,"vramReservedPct":96.81660707509772,"warmupSteps":35,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":71439.09566963676,"meanTps":71363.07681515279,"stepMs":229.3422088623047,"jitter":0.003458127622100906,"achievedTflops":71.2241702930455,"nominalPeakTflops":312.0,"mfuNominalPct":22.828259709309457,"configuredPeakTflops":312.0,"mfuConfiguredPct":22.828259709309457,"vramAllocatedGb":22.055896576,"vramAllocatedPct":25.955979885724556,"vramReservedGb":23.200792576,"vramReservedPct":27.303324685100463,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":74504.14289555869,"meanTps":74490.66961097198,"stepMs":439.81446838378906,"jitter":0.001943490784539282,"achievedTflops":74.27999628760769,"nominalPeakTflops":312.0,"mfuNominalPct":23.807691117822976,"configuredPeakTflops":312.0,"mfuConfiguredPct":23.807691117822976,"vramAllocatedGb":41.987277824,"vramAllocatedPct":49.411772262386975,"vramReservedGb":44.369444864,"vramReservedPct":52.21517132447532,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":53982.76288976166,"meanTps":53949.90248521697,"stepMs":1214.0171508789062,"jitter":0.054450495048957395,"achievedTflops":53.820355099814655,"nominalPeakTflops":312.0,"mfuNominalPct":17.250113814043157,"configuredPeakTflops":312.0,"mfuConfiguredPct":17.250113814043157,"vramAllocatedGb":81.850040832,"vramAllocatedPct":96.32335761824736,"vramReservedGb":82.269175808,"vramReservedPct":96.81660707509772,"warmupSteps":35,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":229883.27011361456,"meanTps":229624.36824214883,"stepMs":285.0838165283203,"jitter":0.003122678507859418,"achievedTflops":229.19166353687854,"nominalPeakTflops":2250.0,"mfuNominalPct":10.186296157194603,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.186296157194603,"vramAllocatedGb":81.850040832,"vramAllocatedPct":42.74086438444696,"vramReservedGb":86.480257024,"vramReservedPct":45.15869387263424,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":256,"tokensPerStep":131072,"status":"complete","stable":true,"tps":240478.49347223298,"meanTps":240439.30179209204,"stepMs":545.0466613769531,"jitter":0.0004390499621810395,"achievedTflops":239.75501103887984,"nominalPeakTflops":2250.0,"mfuNominalPct":10.655778268394661,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.655778268394661,"vramAllocatedGb":161.575566848,"vramAllocatedPct":84.37233897860925,"vramReservedGb":171.102437376,"vramReservedPct":89.34712795985358,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":512,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.499095552,"vramAllocatedPct":99.47577210205387,"vramReservedGb":190.574493696,"vramReservedPct":99.51514388262704,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 192.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 33.81 MiB is free. Including non-PyTorch memory, this process has 178.30 GiB memory in use. Of the allocated memory 177.42 GiB is allocated by PyTorch, and 71.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":230531.6075732813,"meanTps":230605.70679643683,"stepMs":284.2820587158203,"jitter":0.0008865817505925298,"achievedTflops":229.83805046551782,"nominalPeakTflops":2250.0,"mfuNominalPct":10.215024465134125,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.215024465134125,"vramAllocatedGb":81.850040832,"vramAllocatedPct":42.74086438444696,"vramReservedGb":86.480257024,"vramReservedPct":45.15869387263424,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":256,"tokensPerStep":131072,"status":"complete","stable":true,"tps":240349.59890410939,"meanTps":240364.83753863166,"stepMs":545.3389587402344,"jitter":0.0005947103994719096,"achievedTflops":239.62650425161118,"nominalPeakTflops":2250.0,"mfuNominalPct":10.650066855627164,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.650066855627164,"vramAllocatedGb":161.575566848,"vramAllocatedPct":84.37233897860925,"vramReservedGb":171.102437376,"vramReservedPct":89.34712795985358,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":512,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.499095552,"vramAllocatedPct":99.47577210205387,"vramReservedGb":190.574493696,"vramReservedPct":99.51514388262704,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 192.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 33.81 MiB is free. Including non-PyTorch memory, this process has 178.30 GiB memory in use. Of the allocated memory 177.42 GiB is allocated by PyTorch, and 71.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":232702.9696538388,"meanTps":232666.04812960565,"stepMs":281.62940979003906,"jitter":0.0005451046859799863,"achievedTflops":232.0028799771996,"nominalPeakTflops":2250.0,"mfuNominalPct":10.311239110097759,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.311239110097759,"vramAllocatedGb":81.850040832,"vramAllocatedPct":28.476647495287647,"vramReservedGb":86.480257024,"vramReservedPct":30.08755731263508,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":256,"tokensPerStep":131072,"status":"complete","stable":true,"tps":244394.5476441552,"meanTps":244394.42665618192,"stepMs":536.3131103515625,"jitter":0.0001030924481908065,"achievedTflops":243.65928371483318,"nominalPeakTflops":2250.0,"mfuNominalPct":10.82930149843703,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.82930149843703,"vramAllocatedGb":161.575566848,"vramAllocatedPct":56.21414985516938,"vramReservedGb":171.102437376,"vramReservedPct":59.52866663489757,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":512,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":270.793273344,"vramAllocatedPct":94.21234871391006,"vramReservedGb":272.331964416,"vramReservedPct":94.74767848059186,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 31.19 GiB. GPU 0 has a total capacity of 267.69 GiB of which 14.50 GiB is free. Including non-PyTorch memory, this process has 253.18 GiB memory in use. Of the allocated memory 252.20 GiB is allocated by PyTorch, and 167.41 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":232837.24595948405,"meanTps":232552.11269923966,"stepMs":281.4669952392578,"jitter":0.00022118976022243253,"achievedTflops":232.13675231096798,"nominalPeakTflops":2250.0,"mfuNominalPct":10.317188991598575,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.317188991598575,"vramAllocatedGb":81.850040832,"vramAllocatedPct":28.476647495287647,"vramReservedGb":86.480257024,"vramReservedPct":30.08755731263508,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":256,"tokensPerStep":131072,"status":"complete","stable":true,"tps":244458.6603028385,"meanTps":244453.18110513577,"stepMs":536.1724548339844,"jitter":8.117917510539265e-05,"achievedTflops":243.7232034898135,"nominalPeakTflops":2250.0,"mfuNominalPct":10.832142377325043,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.832142377325043,"vramAllocatedGb":161.575566848,"vramAllocatedPct":56.21414985516938,"vramReservedGb":171.102437376,"vramReservedPct":59.52866663489757,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":512,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":270.793273344,"vramAllocatedPct":94.21234871391006,"vramReservedGb":272.331964416,"vramReservedPct":94.74767848059186,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 31.19 GiB. GPU 0 has a total capacity of 267.69 GiB of which 14.50 GiB is free. Including non-PyTorch memory, this process has 253.18 GiB memory in use. Of the allocated memory 252.20 GiB is allocated by PyTorch, and 167.41 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":60317.88110868902,"meanTps":60291.69284146212,"stepMs":67.90689468383789,"jitter":0.00114093018043259,"achievedTflops":60.13641403955895,"nominalPeakTflops":165.2,"mfuNominalPct":36.402187675277816,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":29.261088657639586,"vramAllocatedGb":7.107360256,"vramAllocatedPct":28.135651022952214,"vramReservedGb":7.440695296,"vramReservedPct":29.455212438351758,"warmupSteps":13,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":62587.912219223675,"meanTps":62592.82383884924,"stepMs":130.8878936767578,"jitter":0.0006836691374046398,"achievedTflops":62.399615734257196,"nominalPeakTflops":165.2,"mfuNominalPct":37.77216448804915,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":30.362314038240466,"vramAllocatedGb":12.090205696,"vramAllocatedPct":47.86106177341985,"vramReservedGb":12.6877696,"vramReservedPct":50.226616474641524,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":58114.34988585145,"meanTps":58111.0688577058,"stepMs":281.9269256591797,"jitter":0.00026688843469897187,"achievedTflops":57.93951216684766,"nominalPeakTflops":165.2,"mfuNominalPct":35.07234392666324,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":28.19212334774844,"vramAllocatedGb":22.055896576,"vramAllocatedPct":87.31188327435511,"vramReservedGb":23.265804288,"vramReservedPct":92.10150135035919,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":59997.92828858902,"meanTps":59679.90727714402,"stepMs":68.26902389526367,"jitter":0.0032009214634409373,"achievedTflops":59.81742380135766,"nominalPeakTflops":165.2,"mfuNominalPct":36.209094310749194,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":29.10587485265964,"vramAllocatedGb":7.107360256,"vramAllocatedPct":28.135651022952214,"vramReservedGb":7.440695296,"vramReservedPct":29.455212438351758,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":62507.896080842955,"meanTps":62478.34764736027,"stepMs":131.0554428100586,"jitter":0.001078006664338454,"achievedTflops":62.31984032538899,"nominalPeakTflops":165.2,"mfuNominalPct":37.72387428897639,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":30.32349703611499,"vramAllocatedGb":12.090205696,"vramAllocatedPct":47.86106177341985,"vramReservedGb":12.6877696,"vramReservedPct":50.226616474641524,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":58025.68626873487,"meanTps":58018.05988572752,"stepMs":282.3577117919922,"jitter":0.0006758977216199544,"achievedTflops":57.85111529528715,"nominalPeakTflops":165.2,"mfuNominalPct":35.01883492450796,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":28.149111326877215,"vramAllocatedGb":22.055896576,"vramAllocatedPct":87.31188327435511,"vramReservedGb":23.265804288,"vramReservedPct":92.10150135035919,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":86533.61307442386,"meanTps":86523.57788673873,"stepMs":94.66841506958008,"jitter":0.000280607663604072,"achievedTflops":86.27327566108609,"nominalPeakTflops":209.5,"mfuNominalPct":41.180561174742756,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":32.07819118045496,"vramAllocatedGb":12.090205696,"vramAllocatedPct":35.909164925371684,"vramReservedGb":12.685672448,"vramReservedPct":37.67776294121998,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":93791.74083076662,"meanTps":93769.13922392392,"stepMs":174.6848907470703,"jitter":0.0006499241143699299,"achievedTflops":93.50956725297638,"nominalPeakTflops":209.5,"mfuNominalPct":44.63463830690996,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":34.7687943057386,"vramAllocatedGb":22.055896576,"vramAllocatedPct":65.50830049041744,"vramReservedGb":23.263707136,"vramReservedPct":69.09562312893921,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":31.622397952,"vramAllocatedPct":93.92180182424778,"vramReservedGb":31.725715456,"vramReservedPct":94.22866552732285,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 1.31 GiB is free. Including non-PyTorch memory, this process has 30.04 GiB memory in use. Of the allocated memory 29.36 GiB is allocated by PyTorch, and 94.53 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":86253.70373613323,"meanTps":86241.55447796588,"stepMs":94.97563171386719,"jitter":0.00026695657503635333,"achievedTflops":85.99420843339863,"nominalPeakTflops":209.5,"mfuNominalPct":41.047354860810806,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":31.974428203874304,"vramAllocatedGb":12.090205696,"vramAllocatedPct":35.909164925371684,"vramReservedGb":12.685672448,"vramReservedPct":37.67776294121998,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":93576.74109404092,"meanTps":93578.41498397269,"stepMs":175.08624267578125,"jitter":0.0002913012054954781,"achievedTflops":93.29521434553861,"nominalPeakTflops":209.5,"mfuNominalPct":44.53232188331199,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":34.68909345408801,"vramAllocatedGb":22.055896576,"vramAllocatedPct":65.50830049041744,"vramReservedGb":23.263707136,"vramReservedPct":69.09562312893921,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":31.622397952,"vramAllocatedPct":93.92180182424778,"vramReservedGb":31.725715456,"vramReservedPct":94.22866552732285,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 1.31 GiB is free. Including non-PyTorch memory, this process has 30.04 GiB memory in use. Of the allocated memory 29.36 GiB is allocated by PyTorch, and 94.53 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":140248.208107329,"meanTps":140219.51864162993,"stepMs":116.82145690917969,"jitter":0.0004183362132055453,"achievedTflops":139.8262696902596,"nominalPeakTflops":989.5,"mfuNominalPct":14.131002495225832,"configuredPeakTflops":989.5,"mfuConfiguredPct":14.131002495225832,"vramAllocatedGb":22.10596608,"vramAllocatedPct":26.001667620276212,"vramReservedGb":23.263707136,"vramReservedPct":27.36343566151531,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":151703.33397723854,"meanTps":151542.55016778465,"stepMs":216.00052642822266,"jitter":0.0008499194729966185,"achievedTflops":151.246932676528,"nominalPeakTflops":989.5,"mfuNominalPct":15.285187738911368,"configuredPeakTflops":989.5,"mfuConfiguredPct":15.285187738911368,"vramAllocatedGb":42.037347328,"vramAllocatedPct":49.44552655627536,"vramReservedGb":44.369444864,"vramReservedPct":52.188606174225136,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":134590.90652413314,"meanTps":128281.34825769009,"stepMs":486.9273986816406,"jitter":0.007815234089369167,"achievedTflops":134.1859881667644,"nominalPeakTflops":989.5,"mfuNominalPct":13.560989203311205,"configuredPeakTflops":989.5,"mfuConfiguredPct":13.560989203311205,"vramAllocatedGb":81.900110336,"vramAllocatedPct":96.33324503050265,"vramReservedGb":82.625691648,"vramReservedPct":97.1866944584911,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":140182.1636535921,"meanTps":140150.09906011692,"stepMs":116.87649536132812,"jitter":0.0007664030634425894,"achievedTflops":139.76042393205432,"nominalPeakTflops":989.5,"mfuNominalPct":14.124348047706349,"configuredPeakTflops":989.5,"mfuConfiguredPct":14.124348047706349,"vramAllocatedGb":22.10596608,"vramAllocatedPct":26.001667620276212,"vramReservedGb":23.263707136,"vramReservedPct":27.36343566151531,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":151518.5679040307,"meanTps":151542.48707333626,"stepMs":216.26392364501953,"jitter":0.0011576836426963877,"achievedTflops":151.06272247426858,"nominalPeakTflops":989.5,"mfuNominalPct":15.266571245504656,"configuredPeakTflops":989.5,"mfuConfiguredPct":15.266571245504656,"vramAllocatedGb":42.037347328,"vramAllocatedPct":49.44552655627536,"vramReservedGb":44.369444864,"vramReservedPct":52.188606174225136,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":133804.97310920138,"meanTps":132946.05041933933,"stepMs":489.7874755859375,"jitter":0.008652172623001195,"achievedTflops":133.40241924194268,"nominalPeakTflops":989.5,"mfuNominalPct":13.48180083294014,"configuredPeakTflops":989.5,"mfuConfiguredPct":13.48180083294014,"vramAllocatedGb":81.900110336,"vramAllocatedPct":96.33324503050265,"vramReservedGb":82.625691648,"vramReservedPct":97.1866944584911,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":171707.26802221424,"meanTps":171704.56683055483,"stepMs":190.83641815185547,"jitter":0.0002575387560647209,"achievedTflops":171.1906846458822,"nominalPeakTflops":989.5,"mfuNominalPct":17.300726088517653,"configuredPeakTflops":989.5,"mfuConfiguredPct":17.300726088517653,"vramAllocatedGb":42.037347328,"vramAllocatedPct":28.002354587824215,"vramReservedGb":44.411387904,"vramReservedPct":29.58377516358815,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":180192.39979170522,"meanTps":180062.075395353,"stepMs":363.70013427734375,"jitter":0.0017552008999032196,"achievedTflops":179.6502887946231,"nominalPeakTflops":989.5,"mfuNominalPct":18.155663344580404,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.155663344580404,"vramAllocatedGb":81.900110336,"vramAllocatedPct":54.55615247356549,"vramReservedGb":86.606086144,"vramReservedPct":57.690945975855875,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.508962816,"vramAllocatedPct":90.93276869645923,"vramReservedGb":137.485090816,"vramReservedPct":91.58299722219918,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.60 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.80 GiB is free. Including non-PyTorch memory, this process has 128.00 GiB memory in use. Of the allocated memory 127.13 GiB is allocated by PyTorch, and 150.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":171265.74025883296,"meanTps":171257.81444990417,"stepMs":191.32839965820312,"jitter":0.0005679661083454064,"achievedTflops":170.75048522407536,"nominalPeakTflops":989.5,"mfuNominalPct":17.256239032246118,"configuredPeakTflops":989.5,"mfuConfiguredPct":17.256239032246118,"vramAllocatedGb":42.037347328,"vramAllocatedPct":28.002354587824215,"vramReservedGb":44.411387904,"vramReservedPct":29.58377516358815,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":180502.6980000443,"meanTps":180399.3491528344,"stepMs":363.0749053955078,"jitter":0.001321627224670666,"achievedTflops":179.9596534670788,"nominalPeakTflops":989.5,"mfuNominalPct":18.186928091670417,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.186928091670417,"vramAllocatedGb":81.900110336,"vramAllocatedPct":54.55615247356549,"vramReservedGb":86.606086144,"vramReservedPct":57.690945975855875,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.508962816,"vramAllocatedPct":90.93276869645923,"vramReservedGb":137.485090816,"vramReservedPct":91.58299722219918,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.60 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.80 GiB is free. Including non-PyTorch memory, this process has 128.00 GiB memory in use. Of the allocated memory 127.13 GiB is allocated by PyTorch, and 150.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":54586.56035352127,"meanTps":54570.021875693805,"stepMs":75.03678512573242,"jitter":0.0049642239272394625,"achievedTflops":54.42233603165898,"nominalPeakTflops":364.2,"mfuNominalPct":14.94298078848407,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":12.05544826897024,"vramAllocatedGb":7.107360256,"vramAllocatedPct":13.96973550016166,"vramReservedGb":7.440695296,"vramReservedPct":14.624915788483857,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":60584.6164627444,"meanTps":60538.602985731945,"stepMs":135.2158432006836,"jitter":0.005286595466199281,"achievedTflops":60.402346917100836,"nominalPeakTflops":364.2,"mfuNominalPct":16.58493874714466,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":13.38011966557081,"vramAllocatedGb":12.090205696,"vramAllocatedPct":23.7636716913408,"vramReservedGb":12.6877696,"vramReservedPct":24.93820195048685,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":58822.449896586746,"meanTps":58834.90311712183,"stepMs":278.5331115722656,"jitter":0.0014864217632724452,"achievedTflops":58.64548185020346,"nominalPeakTflops":364.2,"mfuNominalPct":16.102548558540214,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":12.990944972348869,"vramAllocatedGb":22.055896576,"vramAllocatedPct":43.35154407369909,"vramReservedGb":23.265804288,"vramReservedPct":45.72965494854564,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":36756.93133170991,"meanTps":36768.10828996604,"stepMs":445.7390594482422,"jitter":0.0016663795969834951,"achievedTflops":36.64634766271528,"nominalPeakTflops":154.8,"mfuNominalPct":23.673351203304442,"configuredPeakTflops":180.6,"mfuConfiguredPct":20.291443888546667,"vramAllocatedGb":22.055896576,"vramAllocatedPct":43.21810177747728,"vramReservedGb":23.200792576,"vramReservedPct":45.461503295167944,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":37950.43619388316,"meanTps":37947.27616371169,"stepMs":863.4419860839844,"jitter":0.0011390964076364882,"achievedTflops":37.836261851188624,"nominalPeakTflops":154.8,"mfuNominalPct":24.442029619630894,"configuredPeakTflops":180.6,"mfuConfiguredPct":20.950311102540766,"vramAllocatedGb":41.987277824,"vramAllocatedPct":82.27325695439673,"vramReservedGb":44.369444864,"vramReservedPct":86.94106708992751,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.398459904,"vramAllocatedPct":98.75480518571598,"vramReservedGb":50.60427776,"vramReservedPct":99.15810128467888,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 176.25 MiB is free. Process 1415479 has 47.35 GiB memory in use. Of the allocated memory 46.94 GiB is allocated by PyTorch, and 94.28 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":36783.551976316936,"meanTps":36784.5726913912,"stepMs":445.4164733886719,"jitter":0.0010113294956420194,"achievedTflops":36.67288821879356,"nominalPeakTflops":154.8,"mfuNominalPct":23.690496265370516,"configuredPeakTflops":180.6,"mfuConfiguredPct":20.30613965603187,"vramAllocatedGb":22.055896576,"vramAllocatedPct":43.21810177747728,"vramReservedGb":23.200792576,"vramReservedPct":45.461503295167944,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":37996.77777793506,"meanTps":38001.24574252301,"stepMs":862.388916015625,"jitter":0.0010528705718156068,"achievedTflops":37.882464016028806,"nominalPeakTflops":154.8,"mfuNominalPct":24.471875979346773,"configuredPeakTflops":180.6,"mfuConfiguredPct":20.97589369658295,"vramAllocatedGb":41.987277824,"vramAllocatedPct":82.27325695439673,"vramReservedGb":44.369444864,"vramReservedPct":86.94106708992751,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.398459904,"vramAllocatedPct":98.75480518571598,"vramReservedGb":50.60427776,"vramReservedPct":99.15810128467888,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 176.25 MiB is free. Process 1449495 has 47.35 GiB memory in use. Of the allocated memory 46.94 GiB is allocated by PyTorch, and 94.28 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":93304.27147817684,"meanTps":93250.47562012776,"stepMs":87.79876708984375,"jitter":0.0005897310644110685,"achievedTflops":93.0235644577835,"nominalPeakTflops":468.0,"mfuNominalPct":19.8768300123469,"configuredPeakTflops":468.0,"mfuConfiguredPct":19.8768300123469,"vramAllocatedGb":12.090205696,"vramAllocatedPct":11.85617075830383,"vramReservedGb":12.685672448,"vramReservedPct":12.440110822693331,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":105725.43650783492,"meanTps":105660.94095959188,"stepMs":154.96743774414062,"jitter":0.005861089395315808,"achievedTflops":105.40736026339586,"nominalPeakTflops":468.0,"mfuNominalPct":22.522940227221337,"configuredPeakTflops":468.0,"mfuConfiguredPct":22.522940227221337,"vramAllocatedGb":22.055896576,"vramAllocatedPct":21.628951781942018,"vramReservedGb":23.263707136,"vramReservedPct":22.81338227081123,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":97187.33625110268,"meanTps":97182.8615788643,"stepMs":337.16326904296875,"jitter":0.00018740222585359413,"achievedTflops":96.8949469837432,"nominalPeakTflops":468.0,"mfuNominalPct":20.70404850079983,"configuredPeakTflops":468.0,"mfuConfiguredPct":20.70404850079983,"vramAllocatedGb":41.987277824,"vramAllocatedPct":41.17451332712939,"vramReservedGb":44.325404672,"vramReservedPct":43.467380120424245,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":93329.45705658424,"meanTps":93237.66481266663,"stepMs":87.77507400512695,"jitter":0.0004462031658027487,"achievedTflops":93.04867426507603,"nominalPeakTflops":468.0,"mfuNominalPct":19.882195355785477,"configuredPeakTflops":468.0,"mfuConfiguredPct":19.882195355785477,"vramAllocatedGb":12.109866496,"vramAllocatedPct":11.875450976350242,"vramReservedGb":13.665042432,"vramReservedPct":13.40052275097863,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":105249.52600594684,"meanTps":105181.70911660527,"stepMs":155.66815948486328,"jitter":0.00788823995329875,"achievedTflops":104.93288154396365,"nominalPeakTflops":468.0,"mfuNominalPct":22.42155588546232,"configuredPeakTflops":468.0,"mfuConfiguredPct":22.42155588546232,"vramAllocatedGb":22.071756288,"vramAllocatedPct":21.644504491166124,"vramReservedGb":25.394413568,"vramReservedPct":24.902843767894453,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":96998.75429615157,"meanTps":96874.47314377138,"stepMs":337.8187713623047,"jitter":0.0003181598091826879,"achievedTflops":96.70693237986654,"nominalPeakTflops":468.0,"mfuNominalPct":20.663874440142422,"configuredPeakTflops":468.0,"mfuConfiguredPct":20.663874440142422,"vramAllocatedGb":42.0079872,"vramAllocatedPct":41.19482182347161,"vramReservedGb":44.426067968,"vramReservedPct":43.566094836821875,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":68345.60114654717,"meanTps":68292.29325349338,"stepMs":239.72281646728516,"jitter":0.004270296947428291,"achievedTflops":74.58988272145633,"nominalPeakTflops":312.0,"mfuNominalPct":23.907013692774466,"configuredPeakTflops":312.0,"mfuConfiguredPct":23.907013692774466,"vramAllocatedGb":22.062327296,"vramAllocatedPct":25.96354773219117,"vramReservedGb":23.244832768,"vramReservedPct":27.355152382685848,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":71703.69982555918,"meanTps":71709.99486559206,"stepMs":456.99176025390625,"jitter":0.0022612692830163077,"achievedTflops":78.25478847153528,"nominalPeakTflops":312.0,"mfuNominalPct":25.081662971645923,"configuredPeakTflops":312.0,"mfuConfiguredPct":25.081662971645923,"vramAllocatedGb":41.998829056,"vramAllocatedPct":49.425366066856185,"vramReservedGb":44.413485056,"vramReservedPct":52.26699902206071,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":52956.60514085676,"meanTps":52678.757101359035,"stepMs":1237.5415649414062,"jitter":0.03571100994329174,"achievedTflops":57.79489682610734,"nominalPeakTflops":312.0,"mfuNominalPct":18.524005392983124,"configuredPeakTflops":312.0,"mfuConfiguredPct":18.524005392983124,"vramAllocatedGb":81.871832064,"vramAllocatedPct":96.34900213365067,"vramReservedGb":82.227232768,"vramReservedPct":96.76724736311164,"warmupSteps":24,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":68307.76995905416,"meanTps":68357.28440679656,"stepMs":239.85558319091797,"jitter":0.004236148079140665,"achievedTflops":74.54859515077176,"nominalPeakTflops":312.0,"mfuNominalPct":23.893780497042233,"configuredPeakTflops":312.0,"mfuConfiguredPct":23.893780497042233,"vramAllocatedGb":22.062327296,"vramAllocatedPct":25.96354773219117,"vramReservedGb":23.244832768,"vramReservedPct":27.355152382685848,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":71738.55898235565,"meanTps":71738.57983181563,"stepMs":456.7696990966797,"jitter":0.0020213180626639997,"achievedTflops":78.29283247690795,"nominalPeakTflops":312.0,"mfuNominalPct":25.09385656311152,"configuredPeakTflops":312.0,"mfuConfiguredPct":25.09385656311152,"vramAllocatedGb":41.998829056,"vramAllocatedPct":49.425366066856185,"vramReservedGb":44.413485056,"vramReservedPct":52.26699902206071,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":54298.13801101384,"meanTps":53902.57185172159,"stepMs":1206.9658813476562,"jitter":0.020482370601684584,"achievedTflops":59.25899660390337,"nominalPeakTflops":312.0,"mfuNominalPct":18.993268142276722,"configuredPeakTflops":312.0,"mfuConfiguredPct":18.993268142276722,"vramAllocatedGb":81.871832064,"vramAllocatedPct":96.34900213365067,"vramReservedGb":82.143346688,"vramReservedPct":96.66852793913948,"warmupSteps":13,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":221547.01116362997,"meanTps":221382.6561908799,"stepMs":295.81080627441406,"jitter":0.0007695101328740251,"achievedTflops":241.78828341199235,"nominalPeakTflops":2250.0,"mfuNominalPct":10.746145929421882,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.746145929421882,"vramAllocatedGb":81.871832064,"vramAllocatedPct":42.75224343914522,"vramReservedGb":86.522200064,"vramReservedPct":45.18059589939304,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":128,"tokensPerStep":131072,"status":"complete","stable":true,"tps":230938.62916474373,"meanTps":230918.05399965454,"stepMs":567.5620422363281,"jitter":0.000183432944186682,"achievedTflops":252.03795088899258,"nominalPeakTflops":2250.0,"mfuNominalPct":11.20168670617745,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.20168670617745,"vramAllocatedGb":161.617838592,"vramAllocatedPct":84.39441264966955,"vramReservedGb":170.87594496,"vramReservedPct":89.22885701535606,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.299083776,"vramAllocatedPct":99.37132895081763,"vramReservedGb":190.450761728,"vramReservedPct":99.45053290368857,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 192.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 191.81 MiB is free. Including non-PyTorch memory, this process has 178.15 GiB memory in use. Of the allocated memory 177.23 GiB is allocated by PyTorch, and 104.65 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":220748.06543300027,"meanTps":219937.63832012838,"stepMs":296.8814239501953,"jitter":0.0015033708100785851,"achievedTflops":240.91634334052094,"nominalPeakTflops":2250.0,"mfuNominalPct":10.707393037356486,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.707393037356486,"vramAllocatedGb":81.871832064,"vramAllocatedPct":42.75224343914522,"vramReservedGb":86.522200064,"vramReservedPct":45.18059589939304,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":128,"tokensPerStep":131072,"status":"complete","stable":true,"tps":230835.16346613114,"meanTps":230801.88326527923,"stepMs":567.8164367675781,"jitter":0.0002451848305046987,"achievedTflops":251.92503221982088,"nominalPeakTflops":2250.0,"mfuNominalPct":11.196668098658705,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.196668098658705,"vramAllocatedGb":161.617838592,"vramAllocatedPct":84.39441264966955,"vramReservedGb":170.87594496,"vramReservedPct":89.22885701535606,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.299083776,"vramAllocatedPct":99.37132895081763,"vramReservedGb":190.450761728,"vramReservedPct":99.45053290368857,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 192.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 191.81 MiB is free. Including non-PyTorch memory, this process has 178.15 GiB memory in use. Of the allocated memory 177.23 GiB is allocated by PyTorch, and 104.65 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":223827.6771205671,"meanTps":223829.89457208582,"stepMs":292.7966766357422,"jitter":0.0003031071843603141,"achievedTflops":244.27731860081155,"nominalPeakTflops":2250.0,"mfuNominalPct":10.856769715591625,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.856769715591625,"vramAllocatedGb":81.871832064,"vramAllocatedPct":28.484228936003426,"vramReservedGb":86.522200064,"vramReservedPct":30.102149818061097,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":128,"tokensPerStep":131072,"status":"complete","stable":true,"tps":234759.30680159462,"meanTps":234756.6891544699,"stepMs":558.3250427246094,"jitter":0.00016385817150350133,"achievedTflops":256.20769834996133,"nominalPeakTflops":2250.0,"mfuNominalPct":11.387008815553836,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.387008815553836,"vramAllocatedGb":161.617838592,"vramAllocatedPct":56.228856720806384,"vramReservedGb":170.87594496,"vramReservedPct":59.44986710559707,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":270.827354112,"vramAllocatedPct":94.22420583709338,"vramReservedGb":272.961110016,"vramReservedPct":94.96656606198212,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 31.22 GiB. GPU 0 has a total capacity of 267.69 GiB of which 14.46 GiB is free. Including non-PyTorch memory, this process has 253.22 GiB memory in use. Of the allocated memory 252.23 GiB is allocated by PyTorch, and 174.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":223932.66056393817,"meanTps":223909.62602944666,"stepMs":292.65940856933594,"jitter":0.0003075781735579448,"achievedTflops":244.39189368095393,"nominalPeakTflops":2250.0,"mfuNominalPct":10.86186194137573,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.86186194137573,"vramAllocatedGb":81.871832064,"vramAllocatedPct":28.484228936003426,"vramReservedGb":86.522200064,"vramReservedPct":30.102149818061097,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":128,"tokensPerStep":131072,"status":"complete","stable":true,"tps":234639.67326545465,"meanTps":234508.23377236392,"stepMs":558.6097106933594,"jitter":0.0005933794396062656,"achievedTflops":256.0771346958192,"nominalPeakTflops":2250.0,"mfuNominalPct":11.381205986480854,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.381205986480854,"vramAllocatedGb":161.617838592,"vramAllocatedPct":56.228856720806384,"vramReservedGb":170.87594496,"vramReservedPct":59.44986710559707,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":270.827354112,"vramAllocatedPct":94.22420583709338,"vramReservedGb":272.961110016,"vramReservedPct":94.96656606198212,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 31.22 GiB. GPU 0 has a total capacity of 267.69 GiB of which 14.46 GiB is free. Including non-PyTorch memory, this process has 253.22 GiB memory in use. Of the allocated memory 252.23 GiB is allocated by PyTorch, and 174.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":58501.19234374151,"meanTps":58491.00034137759,"stepMs":70.01566696166992,"jitter":0.0011692501008080608,"achievedTflops":63.84605596823385,"nominalPeakTflops":165.2,"mfuNominalPct":38.64773363694543,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":31.06612081821451,"vramAllocatedGb":7.109950976,"vramAllocatedPct":28.145906812892882,"vramReservedGb":7.421820928,"vramReservedPct":29.38049515764568,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":60979.29743477239,"meanTps":60969.7369700377,"stepMs":134.3406753540039,"jitter":0.0008387054832831401,"achievedTflops":66.55056898751505,"nominalPeakTflops":165.2,"mfuNominalPct":40.28484805539652,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":32.38207881964888,"vramAllocatedGb":12.094076416,"vramAllocatedPct":47.876384653252146,"vramReservedGb":12.6877696,"vramReservedPct":50.226616474641524,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":56948.97470910587,"meanTps":56948.99826864181,"stepMs":287.6961364746094,"jitter":0.00033049844880154045,"achievedTflops":62.15202256471761,"nominalPeakTflops":165.2,"mfuNominalPct":37.62228968808572,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":30.241840514824887,"vramAllocatedGb":22.062327296,"vramAllocatedPct":87.33734033397067,"vramReservedGb":23.286775808,"vramReservedPct":92.18452055114372,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":57035.875639756174,"meanTps":56986.75147584072,"stepMs":71.81444931030273,"jitter":0.00666785905266766,"achievedTflops":62.24686305359843,"nominalPeakTflops":165.2,"mfuNominalPct":37.67969918498695,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":30.287987861616305,"vramAllocatedGb":7.109950976,"vramAllocatedPct":28.145906812892882,"vramReservedGb":7.421820928,"vramReservedPct":29.38049515764568,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":60678.97472632938,"meanTps":60650.493541747885,"stepMs":135.00557708740234,"jitter":0.002120076532663677,"achievedTflops":66.22280779695473,"nominalPeakTflops":165.2,"mfuNominalPct":40.08644539767236,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":32.22259725745912,"vramAllocatedGb":12.094076416,"vramAllocatedPct":47.876384653252146,"vramReservedGb":12.6877696,"vramReservedPct":50.226616474641524,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":56869.50289505711,"meanTps":56867.11505958173,"stepMs":288.0981750488281,"jitter":0.00030388868015246385,"achievedTflops":62.06528994125516,"nominalPeakTflops":165.2,"mfuNominalPct":37.56978810003339,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":30.199638281366553,"vramAllocatedGb":22.062327296,"vramAllocatedPct":87.33734033397067,"vramReservedGb":23.286775808,"vramReservedPct":92.18452055114372,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":84384.75102693783,"meanTps":84379.11176452278,"stepMs":97.07915115356445,"jitter":0.00034599627949809806,"achievedTflops":92.09442271321028,"nominalPeakTflops":209.5,"mfuNominalPct":43.959151653083666,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":34.242614248858374,"vramAllocatedGb":12.094076416,"vramAllocatedPct":35.92066136524664,"vramReservedGb":12.685672448,"vramReservedPct":37.67776294121998,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":91354.89197579314,"meanTps":91347.29571322317,"stepMs":179.3445281982422,"jitter":0.0002520925938546255,"achievedTflops":99.70137893578209,"nominalPeakTflops":209.5,"mfuNominalPct":47.59015700991985,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":37.07103816274339,"vramAllocatedGb":22.062327296,"vramAllocatedPct":65.52740039581816,"vramReservedGb":23.244832768,"vramReservedPct":69.03956429831084,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":31.623708672,"vramAllocatedPct":93.92569479859698,"vramReservedGb":31.725715456,"vramReservedPct":94.22866552732285,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 1.31 GiB is free. Including non-PyTorch memory, this process has 30.04 GiB memory in use. Of the allocated memory 29.36 GiB is allocated by PyTorch, and 93.28 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":84090.55801705319,"meanTps":84093.35171656063,"stepMs":97.41878509521484,"jitter":0.0003735854176660042,"achievedTflops":91.77335125086825,"nominalPeakTflops":209.5,"mfuNominalPct":43.80589558513998,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":34.12323322764787,"vramAllocatedGb":12.094076416,"vramAllocatedPct":35.92066136524664,"vramReservedGb":12.685672448,"vramReservedPct":37.67776294121998,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":91300.39226943595,"meanTps":91301.06069299423,"stepMs":179.4515838623047,"jitter":0.0003196394570010779,"achievedTflops":99.64189995487708,"nominalPeakTflops":209.5,"mfuNominalPct":47.56176608824682,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":37.04892264544016,"vramAllocatedGb":22.062327296,"vramAllocatedPct":65.52740039581816,"vramReservedGb":23.244832768,"vramReservedPct":69.03956429831084,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":31.623708672,"vramAllocatedPct":93.92569479859698,"vramReservedGb":31.725715456,"vramReservedPct":94.22866552732285,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 1.31 GiB is free. Including non-PyTorch memory, this process has 30.04 GiB memory in use. Of the allocated memory 29.36 GiB is allocated by PyTorch, and 93.28 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":136376.24864599106,"meanTps":136349.68528776805,"stepMs":120.13822174072266,"jitter":0.0004831446817906696,"achievedTflops":148.83603658244428,"nominalPeakTflops":989.5,"mfuNominalPct":15.041539826421857,"configuredPeakTflops":989.5,"mfuConfiguredPct":15.041539826421857,"vramAllocatedGb":22.1123968,"vramAllocatedPct":26.009231616502117,"vramReservedGb":23.265804288,"vramReservedPct":27.365902391494714,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":147662.42400699187,"meanTps":147640.19604691069,"stepMs":221.91156768798828,"jitter":0.00018232787406812526,"achievedTflops":161.15335448481773,"nominalPeakTflops":989.5,"mfuNominalPct":16.28634203990073,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.28634203990073,"vramAllocatedGb":42.04889856,"vramAllocatedPct":49.45911344471904,"vramReservedGb":44.455428096,"vramReservedPct":52.289742103380654,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":131452.7709552233,"meanTps":129411.93803394657,"stepMs":498.5516815185547,"jitter":0.0029975915036851587,"achievedTflops":143.46273358451424,"nominalPeakTflops":989.5,"mfuNominalPct":14.498507689187898,"configuredPeakTflops":989.5,"mfuConfiguredPct":14.498507689187898,"vramAllocatedGb":81.921901568,"vramAllocatedPct":96.35887649892389,"vramReservedGb":82.290147328,"vramReservedPct":96.79201766178666,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":136229.37611101533,"meanTps":136104.92831237047,"stepMs":120.26774597167969,"jitter":0.0018913250312394257,"achievedTflops":148.67574528388138,"nominalPeakTflops":989.5,"mfuNominalPct":15.025340604737886,"configuredPeakTflops":989.5,"mfuConfiguredPct":15.025340604737886,"vramAllocatedGb":22.1123968,"vramAllocatedPct":26.009231616502117,"vramReservedGb":23.265804288,"vramReservedPct":27.365902391494714,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":147492.29749522518,"meanTps":147469.5434875733,"stepMs":222.16753387451172,"jitter":0.0010729523886774091,"achievedTflops":160.96768464875504,"nominalPeakTflops":989.5,"mfuNominalPct":16.267578034234973,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.267578034234973,"vramAllocatedGb":42.04889856,"vramAllocatedPct":49.45911344471904,"vramReservedGb":44.455428096,"vramReservedPct":52.289742103380654,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":130474.63497582251,"meanTps":129769.2196838545,"stepMs":502.28919982910156,"jitter":0.008462766973413503,"achievedTflops":142.39523184679885,"nominalPeakTflops":989.5,"mfuNominalPct":14.390624744497105,"configuredPeakTflops":989.5,"mfuConfiguredPct":14.390624744497105,"vramAllocatedGb":81.921901568,"vramAllocatedPct":96.35887649892389,"vramReservedGb":82.290147328,"vramReservedPct":96.79201766178666,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":166394.04680570393,"meanTps":166386.88905649175,"stepMs":196.93012237548828,"jitter":0.0001882828952784603,"achievedTflops":181.59636068126082,"nominalPeakTflops":989.5,"mfuNominalPct":18.352335591840404,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.352335591840404,"vramAllocatedGb":42.04889856,"vramAllocatedPct":28.010049214506207,"vramReservedGb":44.36525056,"vramReservedPct":29.553041676616488,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":174631.27397106937,"meanTps":174592.61505094598,"stepMs":375.2821502685547,"jitter":0.002255503543456134,"achievedTflops":190.58616833394603,"nominalPeakTflops":989.5,"mfuNominalPct":19.260855819499348,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.260855819499348,"vramAllocatedGb":81.921901568,"vramAllocatedPct":54.570668275445385,"vramReservedGb":86.669000704,"vramReservedPct":57.732855276271785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.52665856,"vramAllocatedPct":90.94455636931872,"vramReservedGb":137.296347136,"vramReservedPct":91.45726932095147,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.61 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.86 GiB is free. Including non-PyTorch memory, this process has 127.95 GiB memory in use. Of the allocated memory 127.15 GiB is allocated by PyTorch, and 74.03 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":165967.2467423898,"meanTps":165963.32058261632,"stepMs":197.4365463256836,"jitter":0.000416792141307947,"achievedTflops":181.13056674377168,"nominalPeakTflops":989.5,"mfuNominalPct":18.305261924585313,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.305261924585313,"vramAllocatedGb":42.04889856,"vramAllocatedPct":28.010049214506207,"vramReservedGb":44.36525056,"vramReservedPct":29.553041676616488,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":174599.38469813977,"meanTps":174617.33411462753,"stepMs":375.35069274902344,"jitter":0.0017148243567630418,"achievedTflops":190.55136555091408,"nominalPeakTflops":989.5,"mfuNominalPct":19.257338610501677,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.257338610501677,"vramAllocatedGb":81.921901568,"vramAllocatedPct":54.570668275445385,"vramReservedGb":86.669000704,"vramReservedPct":57.732855276271785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.52665856,"vramAllocatedPct":90.94455636931872,"vramReservedGb":137.296347136,"vramReservedPct":91.45726932095147,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.61 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.86 GiB is free. Including non-PyTorch memory, this process has 127.95 GiB memory in use. Of the allocated memory 127.15 GiB is allocated by PyTorch, and 74.03 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":53401.332658743166,"meanTps":53338.532689730506,"stepMs":76.70220565795898,"jitter":0.005793835198776069,"achievedTflops":58.28025613008103,"nominalPeakTflops":364.2,"mfuNominalPct":16.002266922043116,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":12.91004142986087,"vramAllocatedGb":7.109950976,"vramAllocatedPct":13.97482764011959,"vramReservedGb":7.421820928,"vramReservedPct":14.587817636821976,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":59070.32984903296,"meanTps":59061.62657227647,"stepMs":138.68214416503906,"jitter":0.0015294705398686914,"achievedTflops":64.46719177009835,"nominalPeakTflops":364.2,"mfuNominalPct":17.70104112303634,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.280550084768448,"vramAllocatedGb":12.094076416,"vramAllocatedPct":23.771279710724585,"vramReservedGb":12.6877696,"vramReservedPct":24.93820195048685,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":57332.42739843994,"meanTps":57273.27721190401,"stepMs":285.77195739746094,"jitter":0.0042566161858030105,"achievedTflops":62.570508767879794,"nominalPeakTflops":364.2,"mfuNominalPct":17.180260507380503,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":13.860403404505044,"vramAllocatedGb":22.062327296,"vramAllocatedPct":43.36418385193458,"vramReservedGb":23.286775808,"vramReservedPct":45.770875117058836,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":36014.98113873948,"meanTps":35994.748746041514,"stepMs":454.9217987060547,"jitter":0.001648606664943748,"achievedTflops":39.30542967343195,"nominalPeakTflops":154.8,"mfuNominalPct":25.39110444020152,"configuredPeakTflops":180.6,"mfuConfiguredPct":21.76380380588702,"vramAllocatedGb":22.062327296,"vramAllocatedPct":43.23070264865497,"vramReservedGb":23.244832768,"vramReservedPct":45.547799197653575,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":37309.05562912025,"meanTps":37305.001934785025,"stepMs":878.2854309082031,"jitter":0.0005989713567468854,"achievedTflops":40.71773511593381,"nominalPeakTflops":154.8,"mfuNominalPct":26.303446457321584,"configuredPeakTflops":180.6,"mfuConfiguredPct":22.54581124913279,"vramAllocatedGb":41.998829056,"vramAllocatedPct":82.29589136957505,"vramReservedGb":44.413485056,"vramReservedPct":87.02736299241315,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.399770624,"vramAllocatedPct":98.7573735161471,"vramReservedGb":50.62524928,"vramReservedPct":99.1991945715768,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 174.25 MiB is free. Process 1415479 has 47.35 GiB memory in use. Of the allocated memory 46.94 GiB is allocated by PyTorch, and 95.03 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":36096.89175408596,"meanTps":36086.451273235645,"stepMs":453.8894958496094,"jitter":0.0006951937702627389,"achievedTflops":39.39482391519502,"nominalPeakTflops":154.8,"mfuNominalPct":25.448852658394713,"configuredPeakTflops":180.6,"mfuConfiguredPct":21.813302278624043,"vramAllocatedGb":22.062327296,"vramAllocatedPct":43.23070264865497,"vramReservedGb":23.244832768,"vramReservedPct":45.547799197653575,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":37330.19841877289,"meanTps":37327.238672688974,"stepMs":877.7879943847656,"jitter":0.0005920429802404497,"achievedTflops":40.74080958121231,"nominalPeakTflops":154.8,"mfuNominalPct":26.31835244264361,"configuredPeakTflops":180.6,"mfuConfiguredPct":22.55858780798024,"vramAllocatedGb":41.998829056,"vramAllocatedPct":82.29589136957505,"vramReservedGb":44.413485056,"vramReservedPct":87.02736299241315,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.399770624,"vramAllocatedPct":98.7573735161471,"vramReservedGb":50.62524928,"vramReservedPct":99.1991945715768,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 174.25 MiB is free. Process 1449495 has 47.35 GiB memory in use. Of the allocated memory 46.94 GiB is allocated by PyTorch, and 95.03 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":95342.04069926316,"meanTps":95345.27178886146,"stepMs":85.92222213745117,"jitter":0.0003497841479719739,"achievedTflops":104.0528068358593,"nominalPeakTflops":468.0,"mfuNominalPct":22.23350573415797,"configuredPeakTflops":468.0,"mfuConfiguredPct":22.23350573415797,"vramAllocatedGb":12.094076416,"vramAllocatedPct":11.859966551231716,"vramReservedGb":12.685672448,"vramReservedPct":12.440110822693331,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":105472.32132275627,"meanTps":105433.99446851005,"stepMs":155.3393325805664,"jitter":0.000347992324601632,"achievedTflops":115.10862361068864,"nominalPeakTflops":468.0,"mfuNominalPct":24.595859745873643,"configuredPeakTflops":468.0,"mfuConfiguredPct":24.595859745873643,"vramAllocatedGb":22.062327296,"vramAllocatedPct":21.635258019928035,"vramReservedGb":23.244832768,"vramReservedPct":22.79487326148667,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":95946.50295008143,"meanTps":95941.62315895001,"stepMs":341.5236511230469,"jitter":0.00011399981610299733,"achievedTflops":104.71249476956268,"nominalPeakTflops":468.0,"mfuNominalPct":22.374464694351,"configuredPeakTflops":468.0,"mfuConfiguredPct":22.374464694351,"vramAllocatedGb":41.998829056,"vramAllocatedPct":41.185840957320664,"vramReservedGb":44.325404672,"vramReservedPct":43.467380120424245,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":95150.2194364645,"meanTps":95144.68471068745,"stepMs":86.09543991088867,"jitter":0.0006667606582238794,"achievedTflops":103.84346014411017,"nominalPeakTflops":468.0,"mfuNominalPct":22.18877353506628,"configuredPeakTflops":468.0,"mfuConfiguredPct":22.18877353506628,"vramAllocatedGb":12.114605568,"vramAllocatedPct":11.880098312241845,"vramReservedGb":13.679722496,"vramReservedPct":13.414918647119952,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":105180.6298464885,"meanTps":105121.85159431607,"stepMs":155.77012634277344,"jitter":0.0003960454524839029,"achievedTflops":114.79028223039982,"nominalPeakTflops":468.0,"mfuNominalPct":24.527838083418764,"configuredPeakTflops":468.0,"mfuConfiguredPct":24.527838083418764,"vramAllocatedGb":22.07868672,"vramAllocatedPct":21.651300768027486,"vramReservedGb":25.409093632,"vramReservedPct":24.917239664035773,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":95835.2322985623,"meanTps":95831.64760574441,"stepMs":341.92018127441406,"jitter":0.00020795229185399674,"achievedTflops":104.59105806101205,"nominalPeakTflops":468.0,"mfuNominalPct":22.34851667970343,"configuredPeakTflops":468.0,"mfuConfiguredPct":22.34851667970343,"vramAllocatedGb":42.01678592,"vramAllocatedPct":41.20345022313639,"vramReservedGb":44.421873664,"vramReservedPct":43.56198172363864,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":32581.23261802658,"meanTps":32197.716613808232,"stepMs":62.858272552490234,"jitter":0.009748561301945955,"achievedTflops":41.707464203614016,"nominalPeakTflops":312.0,"mfuNominalPct":13.367776988337827,"configuredPeakTflops":312.0,"mfuConfiguredPct":13.367776988337827,"vramAllocatedGb":4.620813312,"vramAllocatedPct":5.437898974937606,"vramReservedGb":4.657774592,"vramReservedPct":5.481396016054246,"warmupSteps":13,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":53779.794120114544,"meanTps":53671.07870202464,"stepMs":76.16243362426758,"jitter":0.004134982596250287,"achievedTflops":68.84389134195582,"nominalPeakTflops":312.0,"mfuNominalPct":22.065349789088405,"configuredPeakTflops":312.0,"mfuConfiguredPct":22.065349789088405,"vramAllocatedGb":7.113212416,"vramAllocatedPct":8.371022132624919,"vramReservedGb":7.430209536,"vramReservedPct":8.744072978334172,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":61456.92067432277,"meanTps":61408.609795712946,"stepMs":133.29662322998047,"jitter":0.0034769887453563538,"achievedTflops":78.67143484530065,"nominalPeakTflops":312.0,"mfuNominalPct":25.2152034760579,"configuredPeakTflops":312.0,"mfuConfiguredPct":25.2152034760579,"vramAllocatedGb":12.097977856,"vramAllocatedPct":14.237229885724554,"vramReservedGb":12.689866752,"vramReservedPct":14.933780861388675,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":66062.76832147548,"meanTps":65917.76156544221,"stepMs":248.00656127929688,"jitter":0.002004608376927402,"achievedTflops":84.56741269620116,"nominalPeakTflops":312.0,"mfuNominalPct":27.104939966731138,"configuredPeakTflops":312.0,"mfuConfiguredPct":27.104939966731138,"vramAllocatedGb":22.067508736,"vramAllocatedPct":25.969645391923827,"vramReservedGb":23.290970112,"vramReservedPct":27.409448065870535,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":68837.75210956861,"meanTps":68807.78343424252,"stepMs":476.0178680419922,"jitter":0.0025197678686775327,"achievedTflops":88.11968889042556,"nominalPeakTflops":312.0,"mfuNominalPct":28.243490028982553,"configuredPeakTflops":312.0,"mfuConfiguredPct":28.243490028982553,"vramAllocatedGb":42.006570496,"vramAllocatedPct":49.43447640432237,"vramReservedGb":44.41767936,"vramReservedPct":52.271934993259315,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":51961.164740334476,"meanTps":51677.2256575974,"stepMs":1261.2496337890625,"jitter":0.02228866964565095,"achievedTflops":66.51585112794466,"nominalPeakTflops":312.0,"mfuNominalPct":21.319183053828418,"configuredPeakTflops":312.0,"mfuConfiguredPct":21.319183053828418,"vramAllocatedGb":81.884694016,"vramAllocatedPct":96.36413842911945,"vramReservedGb":82.25030144,"vramReservedPct":96.79439520470397,"warmupSteps":22,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.967306752,"vramAllocatedPct":98.8150138265037,"vramReservedGb":84.334870528,"vramReservedPct":99.24757289041219,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 384.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 324.75 MiB is free. Process 703773 has 78.81 GiB memory in use. Of the allocated memory 78.20 GiB is allocated by PyTorch, and 110.54 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":30660.76132813464,"meanTps":30895.521427361804,"stepMs":66.79547119140625,"jitter":0.03618680453582267,"achievedTflops":39.249055446760536,"nominalPeakTflops":312.0,"mfuNominalPct":12.5798254637053,"configuredPeakTflops":312.0,"mfuConfiguredPct":12.5798254637053,"vramAllocatedGb":4.620813312,"vramAllocatedPct":5.437898974937606,"vramReservedGb":4.657774592,"vramReservedPct":5.481396016054246,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":53323.76969518188,"meanTps":53048.436301012836,"stepMs":76.81377410888672,"jitter":0.003098059382398031,"achievedTflops":68.26013127978034,"nominalPeakTflops":312.0,"mfuNominalPct":21.8782472050578,"configuredPeakTflops":312.0,"mfuConfiguredPct":21.8782472050578,"vramAllocatedGb":7.113212416,"vramAllocatedPct":8.371022132624919,"vramReservedGb":7.430209536,"vramReservedPct":8.744072978334172,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":61201.93966739521,"meanTps":61098.93451523672,"stepMs":133.85196685791016,"jitter":0.006445383349354731,"achievedTflops":78.3450318714258,"nominalPeakTflops":312.0,"mfuNominalPct":25.1105871382775,"configuredPeakTflops":312.0,"mfuConfiguredPct":25.1105871382775,"vramAllocatedGb":12.097977856,"vramAllocatedPct":14.237229885724554,"vramReservedGb":12.689866752,"vramReservedPct":14.933780861388675,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":65766.72141194555,"meanTps":65764.96314362531,"stepMs":249.12295532226562,"jitter":0.003864435076864503,"achievedTflops":84.18844097261514,"nominalPeakTflops":312.0,"mfuNominalPct":26.983474670709978,"configuredPeakTflops":312.0,"mfuConfiguredPct":26.983474670709978,"vramAllocatedGb":22.067508736,"vramAllocatedPct":25.969645391923827,"vramReservedGb":23.290970112,"vramReservedPct":27.409448065870535,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":68677.4373536276,"meanTps":68701.77608064206,"stepMs":477.12904357910156,"jitter":0.0007406094362270212,"achievedTflops":87.91446884785965,"nominalPeakTflops":312.0,"mfuNominalPct":28.17771437431399,"configuredPeakTflops":312.0,"mfuConfiguredPct":28.17771437431399,"vramAllocatedGb":42.006570496,"vramAllocatedPct":49.43447640432237,"vramReservedGb":44.41767936,"vramReservedPct":52.271934993259315,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":51559.45670540463,"meanTps":49939.45403787176,"stepMs":1271.0762329101562,"jitter":0.03066604329068187,"achievedTflops":66.00162185726106,"nominalPeakTflops":312.0,"mfuNominalPct":21.154365979891363,"configuredPeakTflops":312.0,"mfuConfiguredPct":21.154365979891363,"vramAllocatedGb":81.884694016,"vramAllocatedPct":96.36413842911945,"vramReservedGb":82.25030144,"vramReservedPct":96.79439520470397,"warmupSteps":21,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.967306752,"vramAllocatedPct":98.8150138265037,"vramReservedGb":84.334870528,"vramReservedPct":99.24757289041219,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 384.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 324.75 MiB is free. Process 736416 has 78.81 GiB memory in use. Of the allocated memory 78.20 GiB is allocated by PyTorch, and 110.54 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":16346.801149861396,"meanTps":16392.0296333147,"stepMs":125.28445053100586,"jitter":0.01864390807210654,"achievedTflops":20.925654710320032,"nominalPeakTflops":2250.0,"mfuNominalPct":0.9300290982364459,"configuredPeakTflops":2250.0,"mfuConfiguredPct":0.9300290982364459,"vramAllocatedGb":4.620813312,"vramAllocatedPct":2.412919445201129,"vramReservedGb":4.657774592,"vramReservedPct":2.4322200715648723,"warmupSteps":27,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":32551.4871133176,"meanTps":32325.355806282478,"stepMs":125.83142471313477,"jitter":0.031547342252018606,"achievedTflops":41.66938677457955,"nominalPeakTflops":2250.0,"mfuNominalPct":1.851972745536869,"configuredPeakTflops":2250.0,"mfuConfiguredPct":1.851972745536869,"vramAllocatedGb":7.113212416,"vramAllocatedPct":3.7144128960673544,"vramReservedGb":7.430209536,"vramReservedPct":3.8799440403216314,"warmupSteps":17,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":61643.24630302759,"meanTps":61229.123496954715,"stepMs":132.89371490478516,"jitter":0.017928428623466407,"achievedTflops":78.9099515883757,"nominalPeakTflops":2250.0,"mfuNominalPct":3.507108959483365,"configuredPeakTflops":2250.0,"mfuConfiguredPct":3.507108959483365,"vramAllocatedGb":12.097977856,"vramAllocatedPct":6.317382686841399,"vramReservedGb":12.689866752,"vramReservedPct":6.6264581958753,"warmupSteps":14,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":125223.52484835591,"meanTps":124192.32578691255,"stepMs":130.8380355834961,"jitter":0.01753327301566143,"achievedTflops":160.29951172484252,"nominalPeakTflops":2250.0,"mfuNominalPct":7.124422743326335,"configuredPeakTflops":2250.0,"mfuConfiguredPct":7.124422743326335,"vramAllocatedGb":22.067508736,"vramAllocatedPct":11.523322268389489,"vramReservedGb":23.290970112,"vramReservedPct":12.162195459162302,"warmupSteps":25,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":185945.21003656328,"meanTps":185931.51996036386,"stepMs":176.2239532470703,"jitter":0.004358382079923193,"achievedTflops":238.02976647183658,"nominalPeakTflops":2250.0,"mfuNominalPct":10.579100732081626,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.579100732081626,"vramAllocatedGb":42.006570496,"vramAllocatedPct":21.93520143148567,"vramReservedGb":44.41767936,"vramReservedPct":23.194246337570462,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":204981.4594647867,"meanTps":204711.11964427485,"stepMs":319.7167205810547,"jitter":0.0012903522963706073,"achievedTflops":262.3982027709414,"nominalPeakTflops":2250.0,"mfuNominalPct":11.662142345375173,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.662142345375173,"vramAllocatedGb":81.884694016,"vramAllocatedPct":42.75895975767803,"vramReservedGb":86.50752,"vramReservedPct":45.17293019002746,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":64,"tokensPerStep":131072,"status":"complete","stable":true,"tps":213594.33596222967,"meanTps":213340.605877728,"stepMs":613.6492309570312,"jitter":0.0008574989190900484,"achievedTflops":273.42360633435663,"nominalPeakTflops":2250.0,"mfuNominalPct":12.152160281526962,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.152160281526962,"vramAllocatedGb":161.640940544,"vramAllocatedPct":84.40647614270402,"vramReservedGb":170.798350336,"vramReservedPct":89.18833826585228,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15504.137365609333,"meanTps":15393.261892112554,"stepMs":132.0937728881836,"jitter":0.01916930132302017,"achievedTflops":19.846954894711164,"nominalPeakTflops":2250.0,"mfuNominalPct":0.8820868842093851,"configuredPeakTflops":2250.0,"mfuConfiguredPct":0.8820868842093851,"vramAllocatedGb":4.620813312,"vramAllocatedPct":2.412919445201129,"vramReservedGb":4.657774592,"vramReservedPct":2.4322200715648723,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":30471.282849782907,"meanTps":30361.79606580006,"stepMs":134.42164611816406,"jitter":0.01880720812871894,"achievedTflops":39.00650271875731,"nominalPeakTflops":2250.0,"mfuNominalPct":1.7336223430558804,"configuredPeakTflops":2250.0,"mfuConfiguredPct":1.7336223430558804,"vramAllocatedGb":7.113212416,"vramAllocatedPct":3.7144128960673544,"vramReservedGb":7.430209536,"vramReservedPct":3.8799440403216314,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":58465.92966273219,"meanTps":58061.09913723064,"stepMs":140.11579132080078,"jitter":0.04092120558944533,"achievedTflops":74.84264629049855,"nominalPeakTflops":2250.0,"mfuNominalPct":3.326339835133269,"configuredPeakTflops":2250.0,"mfuConfiguredPct":3.326339835133269,"vramAllocatedGb":12.097977856,"vramAllocatedPct":6.317382686841399,"vramReservedGb":12.689866752,"vramReservedPct":6.6264581958753,"warmupSteps":23,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":121459.42503187238,"meanTps":120454.7443255806,"stepMs":134.89278411865234,"jitter":0.011684845498193024,"achievedTflops":155.48106117095043,"nominalPeakTflops":2250.0,"mfuNominalPct":6.910269385375575,"configuredPeakTflops":2250.0,"mfuConfiguredPct":6.910269385375575,"vramAllocatedGb":22.067508736,"vramAllocatedPct":11.523322268389489,"vramReservedGb":23.290970112,"vramReservedPct":12.162195459162302,"warmupSteps":16,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":185528.67461452508,"meanTps":184074.51365621714,"stepMs":176.61959838867188,"jitter":0.0031007949845450627,"achievedTflops":237.49655655900526,"nominalPeakTflops":2250.0,"mfuNominalPct":10.555402513733567,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.555402513733567,"vramAllocatedGb":42.006570496,"vramAllocatedPct":21.93520143148567,"vramReservedGb":44.41767936,"vramReservedPct":23.194246337570462,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":205000.0976562034,"meanTps":204856.3709805324,"stepMs":319.6876525878906,"jitter":0.001714563966961083,"achievedTflops":262.42206165039033,"nominalPeakTflops":2250.0,"mfuNominalPct":11.663202740017347,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.663202740017347,"vramAllocatedGb":81.884694016,"vramAllocatedPct":42.75895975767803,"vramReservedGb":86.50752,"vramReservedPct":45.17293019002746,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":64,"tokensPerStep":131072,"status":"complete","stable":true,"tps":213555.75202502863,"meanTps":213522.7400019341,"stepMs":613.7601013183594,"jitter":0.0010128159486937226,"achievedTflops":273.3742147659493,"nominalPeakTflops":2250.0,"mfuNominalPct":12.14996510070886,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.14996510070886,"vramAllocatedGb":161.640940544,"vramAllocatedPct":84.40647614270402,"vramReservedGb":170.798350336,"vramReservedPct":89.18833826585228,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":41109.15051672538,"meanTps":41104.086645157404,"stepMs":49.8185920715332,"jitter":0.0018032357234825296,"achievedTflops":52.62411166938698,"nominalPeakTflops":2250.0,"mfuNominalPct":2.33884940752831,"configuredPeakTflops":2250.0,"mfuConfiguredPct":2.33884940752831,"vramAllocatedGb":4.620813312,"vramAllocatedPct":1.607638438414953,"vramReservedGb":4.657774592,"vramReservedPct":1.6204977275592918,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":74756.94575639967,"meanTps":74756.98218994893,"stepMs":54.79089546203613,"jitter":0.0035186075197043924,"achievedTflops":95.69689015944307,"nominalPeakTflops":2250.0,"mfuNominalPct":4.253195118197469,"configuredPeakTflops":2250.0,"mfuConfiguredPct":4.253195118197469,"vramAllocatedGb":7.113212416,"vramAllocatedPct":2.474775094435171,"vramReservedGb":7.430209536,"vramReservedPct":2.585062336219077,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":131592.13994009953,"meanTps":129045.20152520633,"stepMs":62.25295829772949,"jitter":0.0072156108193644685,"achievedTflops":168.45202053503812,"nominalPeakTflops":2250.0,"mfuNominalPct":7.486756468223915,"configuredPeakTflops":2250.0,"mfuConfiguredPct":7.486756468223915,"vramAllocatedGb":12.097977856,"vramAllocatedPct":4.2090370060807425,"vramReservedGb":12.689866752,"vramReservedPct":4.414962516641727,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":168266.21194613012,"meanTps":168198.22911018776,"stepMs":97.3695182800293,"jitter":0.0009730333300881278,"achievedTflops":215.39875712185446,"nominalPeakTflops":2250.0,"mfuNominalPct":9.573278094304642,"configuredPeakTflops":2250.0,"mfuConfiguredPct":9.573278094304642,"vramAllocatedGb":22.067508736,"vramAllocatedPct":7.677560829371886,"vramReservedGb":23.290970112,"vramReservedPct":8.10321826306776,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":191000.6605693633,"meanTps":190346.14424047823,"stepMs":171.5596160888672,"jitter":0.00044959315882455114,"achievedTflops":244.50128412747122,"nominalPeakTflops":2250.0,"mfuNominalPct":10.86672373899872,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.86672373899872,"vramAllocatedGb":42.006570496,"vramAllocatedPct":14.614608475954173,"vramReservedGb":44.41767936,"vramReservedPct":15.453463246152994,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":207595.29523127503,"meanTps":207553.3805384996,"stepMs":315.691162109375,"jitter":0.0003944210333384632,"achievedTflops":265.74419225338414,"nominalPeakTflops":2250.0,"mfuNominalPct":11.810852989039295,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.810852989039295,"vramAllocatedGb":81.884694016,"vramAllocatedPct":28.488703769118747,"vramReservedGb":86.50752,"vramReservedPct":30.097042441161992,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":64,"tokensPerStep":131072,"status":"complete","stable":true,"tps":217479.50504629716,"meanTps":217469.35771391838,"stepMs":602.6866760253906,"jitter":0.00018460395414676822,"achievedTflops":278.39703850613637,"nominalPeakTflops":2250.0,"mfuNominalPct":12.373201711383839,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.373201711383839,"vramAllocatedGb":161.640940544,"vramAllocatedPct":56.23689417731673,"vramReservedGb":170.798350336,"vramReservedPct":59.42287097055894,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":37435.32917889534,"meanTps":37461.83404641554,"stepMs":54.707679748535156,"jitter":0.008076393483431405,"achievedTflops":47.92122722869081,"nominalPeakTflops":2250.0,"mfuNominalPct":2.1298323212751473,"configuredPeakTflops":2250.0,"mfuConfiguredPct":2.1298323212751473,"vramAllocatedGb":4.620813312,"vramAllocatedPct":1.607638438414953,"vramReservedGb":4.657774592,"vramReservedPct":1.6204977275592918,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":73447.28681433486,"meanTps":73416.40482450955,"stepMs":55.767887115478516,"jitter":0.010816156419321226,"achievedTflops":94.02038656961605,"nominalPeakTflops":2250.0,"mfuNominalPct":4.178683847538491,"configuredPeakTflops":2250.0,"mfuConfiguredPct":4.178683847538491,"vramAllocatedGb":7.113212416,"vramAllocatedPct":2.474775094435171,"vramReservedGb":7.430209536,"vramReservedPct":2.585062336219077,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":131578.94918075931,"meanTps":131604.78007153975,"stepMs":62.259199142456055,"jitter":0.0015995233098861637,"achievedTflops":168.43513495156589,"nominalPeakTflops":2250.0,"mfuNominalPct":7.486005997847373,"configuredPeakTflops":2250.0,"mfuConfiguredPct":7.486005997847373,"vramAllocatedGb":12.097977856,"vramAllocatedPct":4.2090370060807425,"vramReservedGb":12.689866752,"vramReservedPct":4.414962516641727,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":168143.72479121733,"meanTps":168151.5098061226,"stepMs":97.44044876098633,"jitter":0.0015049794371773655,"achievedTflops":215.24196045646065,"nominalPeakTflops":2250.0,"mfuNominalPct":9.566309353620474,"configuredPeakTflops":2250.0,"mfuConfiguredPct":9.566309353620474,"vramAllocatedGb":22.067508736,"vramAllocatedPct":7.677560829371886,"vramReservedGb":23.290970112,"vramReservedPct":8.10321826306776,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":190928.76084734785,"meanTps":190899.35683050143,"stepMs":171.6242218017578,"jitter":0.00047519378638843164,"achievedTflops":244.409244789446,"nominalPeakTflops":2250.0,"mfuNominalPct":10.862633101753156,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.862633101753156,"vramAllocatedGb":42.006570496,"vramAllocatedPct":14.614608475954173,"vramReservedGb":44.41767936,"vramReservedPct":15.453463246152994,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":207530.30513951866,"meanTps":207527.07625656185,"stepMs":315.79002380371094,"jitter":0.00024600614188078,"achievedTflops":265.66099798147627,"nominalPeakTflops":2250.0,"mfuNominalPct":11.80715546584339,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.80715546584339,"vramAllocatedGb":81.884694016,"vramAllocatedPct":28.488703769118747,"vramReservedGb":86.50752,"vramReservedPct":30.097042441161992,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":64,"tokensPerStep":131072,"status":"complete","stable":true,"tps":217386.74423746121,"meanTps":217226.3691757127,"stepMs":602.94384765625,"jitter":0.00014017612965034024,"achievedTflops":278.27829474467774,"nominalPeakTflops":2250.0,"mfuNominalPct":12.367924210874566,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.367924210874566,"vramAllocatedGb":161.640940544,"vramAllocatedPct":56.23689417731673,"vramReservedGb":170.798350336,"vramReservedPct":59.42287097055894,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":43908.57655632604,"meanTps":43738.76809924515,"stepMs":46.64236831665039,"jitter":0.003146109930165305,"achievedTflops":56.207676561057475,"nominalPeakTflops":165.2,"mfuNominalPct":34.02401728877572,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":27.349449303896268,"vramAllocatedGb":4.620813312,"vramAllocatedPct":18.29224720653361,"vramReservedGb":4.657774592,"vramReservedPct":18.438564494244435,"warmupSteps":16,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":55393.05321714985,"meanTps":55379.25043883264,"stepMs":73.94429016113281,"jitter":0.00183305284158338,"achievedTflops":70.90903561779047,"nominalPeakTflops":165.2,"mfuNominalPct":42.92314504708867,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":34.5028151574684,"vramAllocatedGb":7.113212416,"vramAllocatedPct":28.158817757936767,"vramReservedGb":7.430209536,"vramReservedPct":29.41370283795949,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":58153.85381956878,"meanTps":58160.768356102126,"stepMs":140.86770629882812,"jitter":0.0011566870914846835,"achievedTflops":74.44315581663756,"nominalPeakTflops":165.2,"mfuNominalPct":45.06244298827939,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":36.222442210674345,"vramAllocatedGb":12.097977856,"vramAllocatedPct":47.89182914324185,"vramReservedGb":12.689866752,"vramReservedPct":50.234918394719976,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":54679.496725825375,"meanTps":54677.85363446727,"stepMs":299.6369934082031,"jitter":0.000306229186622607,"achievedTflops":69.9956069526421,"nominalPeakTflops":165.2,"mfuNominalPct":42.370222126296675,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":34.05836036946336,"vramAllocatedGb":22.067508736,"vramAllocatedPct":87.35785191385202,"vramReservedGb":23.290970112,"vramReservedPct":92.20112439130062,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.593265664,"vramAllocatedPct":97.35647488150306,"vramReservedGb":24.733810688,"vramReservedPct":97.91284540527639,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 23.69 MiB is free. Including non-PyTorch memory, this process has 23.49 GiB memory in use. Of the allocated memory 22.90 GiB is allocated by PyTorch, and 134.03 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":43940.139958033214,"meanTps":43949.30518024121,"stepMs":46.608863830566406,"jitter":0.0005990715352153183,"achievedTflops":56.24808109277914,"nominalPeakTflops":165.2,"mfuNominalPct":34.048475237759774,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":27.369109282026994,"vramAllocatedGb":4.620813312,"vramAllocatedPct":18.29224720653361,"vramReservedGb":4.657774592,"vramReservedPct":18.438564494244435,"warmupSteps":13,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":55348.934660851875,"meanTps":55346.8642356778,"stepMs":74.00323104858398,"jitter":0.001980926624769717,"achievedTflops":70.85255914469059,"nominalPeakTflops":165.2,"mfuNominalPct":42.888958320030625,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":34.47533491753658,"vramAllocatedGb":7.113212416,"vramAllocatedPct":28.158817757936767,"vramReservedGb":7.430209536,"vramReservedPct":29.41370283795949,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":58105.541138116416,"meanTps":57966.62041988344,"stepMs":140.98483276367188,"jitter":0.0013519709333171131,"achievedTflops":74.38131041453512,"nominalPeakTflops":165.2,"mfuNominalPct":45.025006304198016,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":36.19234956509697,"vramAllocatedGb":12.097977856,"vramAllocatedPct":47.89182914324185,"vramReservedGb":12.689866752,"vramReservedPct":50.234918394719976,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":54468.05416055402,"meanTps":54435.818596255755,"stepMs":300.8001708984375,"jitter":0.00459447731575327,"achievedTflops":69.72493784304876,"nominalPeakTflops":165.2,"mfuNominalPct":42.2063788396179,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":33.926658588783766,"vramAllocatedGb":22.067508736,"vramAllocatedPct":87.35785191385202,"vramReservedGb":23.290970112,"vramReservedPct":92.20112439130062,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.593265664,"vramAllocatedPct":97.35647488150306,"vramReservedGb":24.733810688,"vramReservedPct":97.91284540527639,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 23.69 MiB is free. Including non-PyTorch memory, this process has 23.49 GiB memory in use. Of the allocated memory 22.90 GiB is allocated by PyTorch, and 134.03 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":48326.19618429202,"meanTps":48317.2297623874,"stepMs":42.378671646118164,"jitter":0.0008765362435278973,"achievedTflops":61.86270240549507,"nominalPeakTflops":209.5,"mfuNominalPct":29.52873623173989,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":23.001834339740427,"vramAllocatedGb":4.620813312,"vramAllocatedPct":13.724294812027734,"vramReservedGb":4.657774592,"vramReservedPct":13.834073647288738,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":65820.14061373282,"meanTps":65817.93084969337,"stepMs":62.23019218444824,"jitter":0.0005393909492648363,"achievedTflops":84.2568232671848,"nominalPeakTflops":209.5,"mfuNominalPct":40.218054065482,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":31.328432406347954,"vramAllocatedGb":7.113212416,"vramAllocatedPct":21.126978665040703,"vramReservedGb":7.430209536,"vramReservedPct":22.068492990699685,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":80238.83623620162,"meanTps":80231.12448485201,"stepMs":102.09519958496094,"jitter":0.000415414933781065,"achievedTflops":102.71429657972602,"nominalPeakTflops":209.5,"mfuNominalPct":49.02830385667114,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":38.19130336019658,"vramAllocatedGb":12.097977856,"vramAllocatedPct":35.93224904670791,"vramReservedGb":12.689866752,"vramReservedPct":37.69022045913739,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":87036.85494762997,"meanTps":86996.65491162425,"stepMs":188.24209594726562,"jitter":0.0012008250548731527,"achievedTflops":111.4164879727406,"nominalPeakTflops":209.5,"mfuNominalPct":53.18209449772821,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":41.42695840001029,"vramAllocatedGb":22.067508736,"vramAllocatedPct":65.54278981004232,"vramReservedGb":23.290970112,"vramReservedPct":69.17659699540239,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":31.626330112,"vramAllocatedPct":93.93348074729536,"vramReservedGb":31.822184448,"vramReservedPct":94.51518843942337,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 1.24 GiB is free. Including non-PyTorch memory, this process has 30.11 GiB memory in use. Of the allocated memory 29.36 GiB is allocated by PyTorch, and 162.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":48125.2575869402,"meanTps":48116.82764188972,"stepMs":42.55561637878418,"jitter":0.000839467117234558,"achievedTflops":61.60547949884731,"nominalPeakTflops":209.5,"mfuNominalPct":29.405956801359096,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":22.90619353426257,"vramAllocatedGb":4.620813312,"vramAllocatedPct":13.724294812027734,"vramReservedGb":4.657774592,"vramReservedPct":13.834073647288738,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":65613.28302831302,"meanTps":65603.96280646426,"stepMs":62.42638397216797,"jitter":0.0008078573138698156,"achievedTflops":83.99202342243095,"nominalPeakTflops":209.5,"mfuNominalPct":40.091657958200926,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":31.229974338313795,"vramAllocatedGb":7.113212416,"vramAllocatedPct":21.126978665040703,"vramReservedGb":7.430209536,"vramReservedPct":22.068492990699685,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":80005.72424359554,"meanTps":79983.00347303186,"stepMs":102.39267349243164,"jitter":0.0008402916663950115,"achievedTflops":102.41588828434215,"nominalPeakTflops":209.5,"mfuNominalPct":48.885865529518924,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":38.080348973965066,"vramAllocatedGb":12.097977856,"vramAllocatedPct":35.93224904670791,"vramReservedGb":12.689866752,"vramReservedPct":37.69022045913739,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":86989.83659315798,"meanTps":86993.16209905068,"stepMs":188.34384155273438,"jitter":0.00013338629323012593,"achievedTflops":111.35629944767638,"nominalPeakTflops":209.5,"mfuNominalPct":53.1533648914923,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":41.40457905948931,"vramAllocatedGb":22.067508736,"vramAllocatedPct":65.54278981004232,"vramReservedGb":23.290970112,"vramReservedPct":69.17659699540239,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":31.626330112,"vramAllocatedPct":93.93348074729536,"vramReservedGb":31.822184448,"vramReservedPct":94.51518843942337,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 1.24 GiB is free. Including non-PyTorch memory, this process has 30.11 GiB memory in use. Of the allocated memory 29.36 GiB is allocated by PyTorch, and 162.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":41490.15871058076,"meanTps":41303.19986894275,"stepMs":49.36110305786133,"jitter":0.01150924929038247,"achievedTflops":53.11184292844668,"nominalPeakTflops":989.5,"mfuNominalPct":5.367543499590367,"configuredPeakTflops":989.5,"mfuConfiguredPct":5.367543499590367,"vramAllocatedGb":4.670882816,"vramAllocatedPct":5.494025551082895,"vramReservedGb":4.720689152,"vramReservedPct":5.552609183635713,"warmupSteps":24,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":81185.15149132305,"meanTps":80988.36685177285,"stepMs":50.45257568359375,"jitter":0.004295567669047863,"achievedTflops":103.92568136958428,"nominalPeakTflops":989.5,"mfuNominalPct":10.502848041393056,"configuredPeakTflops":989.5,"mfuConfiguredPct":10.502848041393056,"vramAllocatedGb":7.16328192,"vramAllocatedPct":8.425656444062273,"vramReservedGb":7.493124096,"vramReservedPct":8.81362621640622,"warmupSteps":14,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":115030.28161073617,"meanTps":115025.01798461317,"stepMs":71.21603012084961,"jitter":0.0015412202944422675,"achievedTflops":147.25106962211692,"nominalPeakTflops":989.5,"mfuNominalPct":14.881361255393323,"configuredPeakTflops":989.5,"mfuConfiguredPct":14.881361255393323,"vramAllocatedGb":12.14804736,"vramAllocatedPct":14.288879687365101,"vramReservedGb":12.731809792,"vramReservedPct":14.975517704954427,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":129285.90698833123,"meanTps":129265.06664068224,"stepMs":126.72688293457031,"jitter":0.0008795892451608172,"achievedTflops":165.499795571399,"nominalPeakTflops":989.5,"mfuNominalPct":16.725598339706824,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.725598339706824,"vramAllocatedGb":22.11757824,"vramAllocatedPct":26.015326173970756,"vramReservedGb":23.353884672,"vramReservedPct":27.469505050629632,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":139837.54549801396,"meanTps":139751.53640576234,"stepMs":234.3290557861328,"jitter":0.00025936105704054357,"achievedTflops":179.00702197352652,"nominalPeakTflops":989.5,"mfuNominalPct":18.090654065035526,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.090654065035526,"vramAllocatedGb":42.05664,"vramAllocatedPct":49.46821914718207,"vramReservedGb":44.4596224,"vramReservedPct":52.29467556333946,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":127396.74692376782,"meanTps":126780.75990764055,"stepMs":514.4244384765625,"jitter":0.0029573440052774393,"achievedTflops":163.08146853351764,"nominalPeakTflops":989.5,"mfuNominalPct":16.48119944755105,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.48119944755105,"vramAllocatedGb":81.93476352,"vramAllocatedPct":96.3740050936047,"vramReservedGb":82.39710208,"vramReservedPct":96.9178208907362,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.017376256,"vramAllocatedPct":98.82363357612637,"vramReservedGb":84.188069888,"vramReservedPct":99.0244082931462,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 384.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 120.19 MiB is free. Including non-PyTorch memory, this process has 79.05 GiB memory in use. Of the allocated memory 78.25 GiB is allocated by PyTorch, and 82.79 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":40646.39153067735,"meanTps":40601.166222599015,"stepMs":50.38577651977539,"jitter":0.002773912771252139,"achievedTflops":52.03173064833192,"nominalPeakTflops":989.5,"mfuNominalPct":5.2583861190835695,"configuredPeakTflops":989.5,"mfuConfiguredPct":5.2583861190835695,"vramAllocatedGb":4.670882816,"vramAllocatedPct":5.494025551082895,"vramReservedGb":4.720689152,"vramReservedPct":5.552609183635713,"warmupSteps":24,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":81218.92044993392,"meanTps":80557.55692018992,"stepMs":50.43159866333008,"jitter":0.006106371333386938,"achievedTflops":103.96890925015494,"nominalPeakTflops":989.5,"mfuNominalPct":10.507216700369373,"configuredPeakTflops":989.5,"mfuConfiguredPct":10.507216700369373,"vramAllocatedGb":7.16328192,"vramAllocatedPct":8.425656444062273,"vramReservedGb":7.493124096,"vramReservedPct":8.81362621640622,"warmupSteps":16,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":114814.2245187737,"meanTps":114820.42718378993,"stepMs":71.35004425048828,"jitter":0.0012610511684514733,"achievedTflops":146.9744934245676,"nominalPeakTflops":989.5,"mfuNominalPct":14.853410149021487,"configuredPeakTflops":989.5,"mfuConfiguredPct":14.853410149021487,"vramAllocatedGb":12.14804736,"vramAllocatedPct":14.288879687365101,"vramReservedGb":12.731809792,"vramReservedPct":14.975517704954427,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":129067.32464357538,"meanTps":128996.99623626849,"stepMs":126.94150161743164,"jitter":0.00110960138087162,"achievedTflops":165.21998678005193,"nominalPeakTflops":989.5,"mfuNominalPct":16.69732054371419,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.69732054371419,"vramAllocatedGb":22.11757824,"vramAllocatedPct":26.015326173970756,"vramReservedGb":23.353884672,"vramReservedPct":27.469505050629632,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":139574.54222857492,"meanTps":139571.95143566234,"stepMs":234.7706069946289,"jitter":0.0002916359245574194,"achievedTflops":178.67034964519073,"nominalPeakTflops":989.5,"mfuNominalPct":18.056629575057173,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.056629575057173,"vramAllocatedGb":42.05664,"vramAllocatedPct":49.46821914718207,"vramReservedGb":44.4596224,"vramReservedPct":52.29467556333946,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":124184.46173678453,"meanTps":124032.72139213748,"stepMs":527.7310791015625,"jitter":0.0038457578264853146,"achievedTflops":158.96939975396575,"nominalPeakTflops":989.5,"mfuNominalPct":16.06562908074439,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.06562908074439,"vramAllocatedGb":81.93476352,"vramAllocatedPct":96.3740050936047,"vramReservedGb":82.39710208,"vramReservedPct":96.9178208907362,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.017376256,"vramAllocatedPct":98.82363357612637,"vramReservedGb":84.188069888,"vramReservedPct":99.0244082931462,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 384.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 120.19 MiB is free. Including non-PyTorch memory, this process has 79.05 GiB memory in use. Of the allocated memory 78.25 GiB is allocated by PyTorch, and 82.79 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":38038.6665679927,"meanTps":37618.324290677665,"stepMs":53.83995246887207,"jitter":0.019449576944810937,"achievedTflops":48.69356364866277,"nominalPeakTflops":989.5,"mfuNominalPct":4.921027149940653,"configuredPeakTflops":989.5,"mfuConfiguredPct":4.921027149940653,"vramAllocatedGb":4.670882816,"vramAllocatedPct":3.111416993828419,"vramReservedGb":4.720689152,"vramReservedPct":3.1445945078734914,"warmupSteps":31,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":75325.51209461587,"meanTps":75289.59809608669,"stepMs":54.37732696533203,"jitter":0.002860008516495273,"achievedTflops":96.42471591350655,"nominalPeakTflops":989.5,"mfuNominalPct":9.744791906367514,"configuredPeakTflops":989.5,"mfuConfiguredPct":9.744791906367514,"vramAllocatedGb":7.16328192,"vramAllocatedPct":4.771679782058541,"vramReservedGb":7.493124096,"vramReservedPct":4.991397679534423,"warmupSteps":29,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":123833.07831309635,"meanTps":123829.7822333338,"stepMs":66.15356826782227,"jitter":0.0015895722354840433,"achievedTflops":158.51959137080746,"nominalPeakTflops":989.5,"mfuNominalPct":16.020170931865334,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.020170931865334,"vramAllocatedGb":12.14804736,"vramAllocatedPct":8.092183530758152,"vramReservedGb":12.731809792,"vramReservedPct":8.481045427498875,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":142895.97945210038,"meanTps":142881.32746792646,"stepMs":114.65682983398438,"jitter":0.0004574086822860869,"achievedTflops":182.9221447116577,"nominalPeakTflops":989.5,"mfuNominalPct":18.48632083998562,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.48632083998562,"vramAllocatedGb":22.11757824,"vramAllocatedPct":14.733191028157375,"vramReservedGb":23.353884672,"vramReservedPct":15.556732314384364,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":156019.95675725944,"meanTps":156011.77031499852,"stepMs":210.0244140625,"jitter":0.0003179194298816637,"achievedTflops":199.72224003282474,"nominalPeakTflops":989.5,"mfuNominalPct":20.184157658698812,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.184157658698812,"vramAllocatedGb":42.05664,"vramAllocatedPct":28.01520602295582,"vramReservedGb":44.4596224,"vramReservedPct":29.61590562724035,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":163992.31021029243,"meanTps":163919.6436965116,"stepMs":399.6284942626953,"jitter":0.001205777628628701,"achievedTflops":209.92770555830532,"nominalPeakTflops":989.5,"mfuNominalPct":21.215533659252685,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.215533659252685,"vramAllocatedGb":81.93476352,"vramAllocatedPct":54.579236012552705,"vramReservedGb":86.54946304,"vramReservedPct":57.65322760548156,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.537472512,"vramAllocatedPct":90.95175987138647,"vramReservedGb":137.298444288,"vramReservedPct":91.45866629763199,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.80 GiB is free. Including non-PyTorch memory, this process has 128.01 GiB memory in use. Of the allocated memory 127.16 GiB is allocated by PyTorch, and 125.72 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":37197.699022023044,"meanTps":37089.381932202916,"stepMs":55.05716896057129,"jitter":0.0031543333423520756,"achievedTflops":47.617035199566445,"nominalPeakTflops":989.5,"mfuNominalPct":4.812231955489282,"configuredPeakTflops":989.5,"mfuConfiguredPct":4.812231955489282,"vramAllocatedGb":4.670882816,"vramAllocatedPct":3.111416993828419,"vramReservedGb":4.720689152,"vramReservedPct":3.1445945078734914,"warmupSteps":44,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":75105.69079190574,"meanTps":74907.85114799833,"stepMs":54.53647994995117,"jitter":0.013371742502366502,"achievedTflops":96.14332112339962,"nominalPeakTflops":989.5,"mfuNominalPct":9.716353827529016,"configuredPeakTflops":989.5,"mfuConfiguredPct":9.716353827529016,"vramAllocatedGb":7.16328192,"vramAllocatedPct":4.771679782058541,"vramReservedGb":7.493124096,"vramReservedPct":4.991397679534423,"warmupSteps":26,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":124355.09437624631,"meanTps":124055.12875131414,"stepMs":65.87586975097656,"jitter":0.0023488643209864817,"achievedTflops":159.1878277915343,"nominalPeakTflops":989.5,"mfuNominalPct":16.0877036676639,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.0877036676639,"vramAllocatedGb":12.14804736,"vramAllocatedPct":8.092183530758152,"vramReservedGb":12.731809792,"vramReservedPct":8.481045427498875,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":143347.3983734302,"meanTps":143346.57643281968,"stepMs":114.29576110839844,"jitter":0.0003704774536021489,"achievedTflops":183.50000923639578,"nominalPeakTflops":989.5,"mfuNominalPct":18.54472048877168,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.54472048877168,"vramAllocatedGb":22.11757824,"vramAllocatedPct":14.733191028157375,"vramReservedGb":23.353884672,"vramReservedPct":15.556732314384364,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":156420.04261868593,"meanTps":156417.08037994453,"stepMs":209.48722076416016,"jitter":0.0002630654504682057,"achievedTflops":200.23439274783854,"nominalPeakTflops":989.5,"mfuNominalPct":20.235916396951847,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.235916396951847,"vramAllocatedGb":42.05664,"vramAllocatedPct":28.01520602295582,"vramReservedGb":44.4596224,"vramReservedPct":29.61590562724035,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":163723.83242480844,"meanTps":163708.64851474474,"stepMs":400.2838134765625,"jitter":0.0011781450691876643,"achievedTflops":209.58402526361493,"nominalPeakTflops":989.5,"mfuNominalPct":21.180800936191503,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.180800936191503,"vramAllocatedGb":81.93476352,"vramAllocatedPct":54.579236012552705,"vramReservedGb":86.54946304,"vramReservedPct":57.65322760548156,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.537472512,"vramAllocatedPct":90.95175987138647,"vramReservedGb":137.298444288,"vramReservedPct":91.45866629763199,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.80 GiB is free. Including non-PyTorch memory, this process has 128.01 GiB memory in use. Of the allocated memory 127.16 GiB is allocated by PyTorch, and 125.72 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":31636.871769441983,"meanTps":31441.118681617518,"stepMs":64.73459243774414,"jitter":0.008983691319601278,"achievedTflops":40.498581263260036,"nominalPeakTflops":364.2,"mfuNominalPct":11.119874042630432,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":8.971106111687353,"vramAllocatedGb":4.620813312,"vramAllocatedPct":9.082350892481054,"vramReservedGb":4.657774592,"vramReservedPct":9.154999426782032,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":52999.69051400244,"meanTps":52962.945006622685,"stepMs":77.2834701538086,"jitter":0.007053243664989453,"achievedTflops":67.84527524880563,"nominalPeakTflops":364.2,"mfuNominalPct":18.62857640000155,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.02885149178869,"vramAllocatedGb":7.113212416,"vramAllocatedPct":13.981238100896668,"vramReservedGb":7.430209536,"vramReservedPct":14.604305704227258,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":57436.297191866564,"meanTps":57368.18535234151,"stepMs":142.62757873535156,"jitter":0.0037731807813663944,"achievedTflops":73.5246065488791,"nominalPeakTflops":364.2,"mfuNominalPct":20.18797543901129,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.286917383163697,"vramAllocatedGb":12.097977856,"vramAllocatedPct":23.77894811121459,"vramReservedGb":12.689866752,"vramReservedPct":24.94232396733817,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":55391.41582568659,"meanTps":55385.76342561257,"stepMs":295.78590393066406,"jitter":0.004644601079463751,"achievedTflops":70.90693958150352,"nominalPeakTflops":364.2,"mfuNominalPct":19.469231076744514,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.707060820368763,"vramAllocatedGb":22.067508736,"vramAllocatedPct":43.37436813185044,"vramReservedGb":23.290970112,"vramReservedPct":45.77911915076148,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":51540.15511460794,"meanTps":51548.34866734153,"stepMs":635.776123046875,"jitter":0.0026154053508664473,"achievedTflops":65.97691375561666,"nominalPeakTflops":364.2,"mfuNominalPct":18.115572145968333,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.614978494573485,"vramAllocatedGb":42.006570496,"vramAllocatedPct":82.56520817312213,"vramReservedGb":44.41767936,"vramReservedPct":87.30431691096057,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":30617.01457269785,"meanTps":30605.752526319953,"stepMs":66.89091110229492,"jitter":0.006679078777700732,"achievedTflops":39.19305491854864,"nominalPeakTflops":364.2,"mfuNominalPct":10.761409917229171,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":8.681910416315322,"vramAllocatedGb":4.620813312,"vramAllocatedPct":9.082350892481054,"vramReservedGb":4.657774592,"vramReservedPct":9.154999426782032,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":52181.75962558507,"meanTps":51670.244480387264,"stepMs":78.4948616027832,"jitter":0.0026936442059921684,"achievedTflops":66.79823618648291,"nominalPeakTflops":364.2,"mfuNominalPct":18.34108626756807,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.796915008134626,"vramAllocatedGb":7.113212416,"vramAllocatedPct":13.981238100896668,"vramReservedGb":7.430209536,"vramReservedPct":14.604305704227258,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":56970.174089443884,"meanTps":56885.80956821252,"stepMs":143.79454040527344,"jitter":0.005838147561488197,"achievedTflops":72.92791909887714,"nominalPeakTflops":364.2,"mfuNominalPct":20.02414033467247,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.154741236185036,"vramAllocatedGb":12.097977856,"vramAllocatedPct":23.77894811121459,"vramReservedGb":12.689866752,"vramReservedPct":24.94232396733817,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":55071.554278039635,"meanTps":55137.792150554444,"stepMs":297.5038604736328,"jitter":0.003914319323484849,"achievedTflops":70.49748257277092,"nominalPeakTflops":364.2,"mfuNominalPct":19.356804660288557,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.61635931530526,"vramAllocatedGb":22.067508736,"vramAllocatedPct":43.37436813185044,"vramReservedGb":23.290970112,"vramReservedPct":45.77911915076148,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":51303.7228844919,"meanTps":51355.65390046184,"stepMs":638.7060852050781,"jitter":0.0019770483146425498,"achievedTflops":65.67425520092806,"nominalPeakTflops":364.2,"mfuNominalPct":18.03246985198464,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.547934614886158,"vramAllocatedGb":42.006570496,"vramAllocatedPct":82.56520817312213,"vramReservedGb":44.41767936,"vramReservedPct":87.30431691096057,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":23340.069323196523,"meanTps":23121.19185956272,"stepMs":87.74609756469727,"jitter":0.01413709713538056,"achievedTflops":29.877786307829687,"nominalPeakTflops":154.8,"mfuNominalPct":19.300895547693596,"configuredPeakTflops":180.6,"mfuConfiguredPct":16.54362475516594,"vramAllocatedGb":4.620813312,"vramAllocatedPct":9.054394108378407,"vramReservedGb":4.657774592,"vramReservedPct":9.12681902002784,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":28641.081346091727,"meanTps":28622.27318958245,"stepMs":143.01136016845703,"jitter":0.004826790363293043,"achievedTflops":36.66364894783009,"nominalPeakTflops":154.8,"mfuNominalPct":23.684527744076284,"configuredPeakTflops":180.6,"mfuConfiguredPct":20.301023780636815,"vramAllocatedGb":7.113212416,"vramAllocatedPct":13.938201836420467,"vramReservedGb":7.430209536,"vramReservedPct":14.559351547932751,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":32730.02106960664,"meanTps":32607.978296936173,"stepMs":250.29009246826172,"jitter":0.0024464283829832103,"achievedTflops":41.89792934319111,"nominalPeakTflops":154.8,"mfuNominalPct":27.065845828934822,"configuredPeakTflops":180.6,"mfuConfiguredPct":23.19929642480128,"vramAllocatedGb":12.097977856,"vramAllocatedPct":23.705753084243806,"vramReservedGb":12.689866752,"vramReservedPct":24.86554790193087,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":35144.09381291005,"meanTps":35114.875506754026,"stepMs":466.1949768066406,"jitter":0.0026929626570662475,"achievedTflops":44.988200779715584,"nominalPeakTflops":154.8,"mfuNominalPct":29.06214520653461,"configuredPeakTflops":180.6,"mfuConfiguredPct":24.910410177029668,"vramAllocatedGb":22.067508736,"vramAllocatedPct":43.240855579890486,"vramReservedGb":23.290970112,"vramReservedPct":45.638204428829,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":36261.35580916241,"meanTps":36268.247357065826,"stepMs":903.6617431640625,"jitter":0.0014600095072173395,"achievedTflops":46.418415690890306,"nominalPeakTflops":154.8,"mfuNominalPct":29.986056647861954,"configuredPeakTflops":180.6,"mfuConfiguredPct":25.702334269595962,"vramAllocatedGb":42.006570496,"vramAllocatedPct":82.31106057118384,"vramReservedGb":44.41767936,"vramReservedPct":87.03558164979273,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.402392064,"vramAllocatedPct":98.76251017700933,"vramReservedGb":50.633637888,"vramReservedPct":99.21563188633597,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 144.25 MiB is free. Process 1348507 has 47.38 GiB memory in use. Of the allocated memory 46.94 GiB is allocated by PyTorch, and 120.53 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":23212.95976819162,"meanTps":23056.808470457592,"stepMs":88.22657775878906,"jitter":0.0058977536645084235,"achievedTflops":29.71507247568415,"nominalPeakTflops":154.8,"mfuNominalPct":19.195783253025937,"configuredPeakTflops":180.6,"mfuConfiguredPct":16.45352850259366,"vramAllocatedGb":4.620813312,"vramAllocatedPct":9.054394108378407,"vramReservedGb":4.657774592,"vramReservedPct":9.12681902002784,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":28486.33245207866,"meanTps":28464.098374878602,"stepMs":143.7882537841797,"jitter":0.003210071347564077,"achievedTflops":36.46555380412371,"nominalPeakTflops":154.8,"mfuNominalPct":23.556559304989474,"configuredPeakTflops":180.6,"mfuConfiguredPct":20.191336547133837,"vramAllocatedGb":7.113212416,"vramAllocatedPct":13.938201836420467,"vramReservedGb":7.430209536,"vramReservedPct":14.559351547932751,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":32554.2860003195,"meanTps":32541.43137643581,"stepMs":251.6412124633789,"jitter":0.0018758646491779415,"achievedTflops":41.67296964944528,"nominalPeakTflops":154.8,"mfuNominalPct":26.920523029357412,"configuredPeakTflops":180.6,"mfuConfiguredPct":23.074734025163497,"vramAllocatedGb":12.097977856,"vramAllocatedPct":23.705753084243806,"vramReservedGb":12.689866752,"vramReservedPct":24.86554790193087,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":34838.37867196874,"meanTps":34833.01052497281,"stepMs":470.28594970703125,"jitter":0.0016744554918681168,"achievedTflops":44.59685268534491,"nominalPeakTflops":154.8,"mfuNominalPct":28.80933636004193,"configuredPeakTflops":180.6,"mfuConfiguredPct":24.693716880035943,"vramAllocatedGb":22.067508736,"vramAllocatedPct":43.240855579890486,"vramReservedGb":23.290970112,"vramReservedPct":45.638204428829,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":36046.42257303758,"meanTps":36049.66583577052,"stepMs":909.0499877929688,"jitter":0.0006149845355892942,"achievedTflops":46.14327814907479,"nominalPeakTflops":154.8,"mfuNominalPct":29.80831921774857,"configuredPeakTflops":180.6,"mfuConfiguredPct":25.549987900927345,"vramAllocatedGb":42.006570496,"vramAllocatedPct":82.31106057118384,"vramReservedGb":44.41767936,"vramReservedPct":87.03558164979273,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.402392064,"vramAllocatedPct":98.76251017700933,"vramReservedGb":50.633637888,"vramReservedPct":99.21563188633597,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 144.25 MiB is free. Process 1380474 has 47.38 GiB memory in use. Of the allocated memory 46.94 GiB is allocated by PyTorch, and 120.53 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":44877.41836792909,"meanTps":44825.126488390255,"stepMs":45.63542366027832,"jitter":0.0015677344891698837,"achievedTflops":57.447897753733955,"nominalPeakTflops":468.0,"mfuNominalPct":12.27519182772093,"configuredPeakTflops":468.0,"mfuConfiguredPct":12.27519182772093,"vramAllocatedGb":4.620813312,"vramAllocatedPct":4.531366384232895,"vramReservedGb":4.657774592,"vramReservedPct":4.567612189982127,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":68119.25563151488,"meanTps":68083.70240819834,"stepMs":60.12984085083008,"jitter":0.0012730699867795298,"achievedTflops":87.19993651364565,"nominalPeakTflops":468.0,"mfuNominalPct":18.632465067018305,"configuredPeakTflops":468.0,"mfuConfiguredPct":18.632465067018305,"vramAllocatedGb":7.113212416,"vramAllocatedPct":6.9755191238876115,"vramReservedGb":7.430209536,"vramReservedPct":7.28638000410026,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":89557.55345834287,"meanTps":89937.550574814,"stepMs":91.47190475463867,"jitter":0.009564975402658622,"achievedTflops":114.64325180135933,"nominalPeakTflops":468.0,"mfuNominalPct":24.49642132507678,"configuredPeakTflops":468.0,"mfuConfiguredPct":24.49642132507678,"vramAllocatedGb":12.097977856,"vramAllocatedPct":11.863792469500302,"vramReservedGb":12.689866752,"vramReservedPct":12.444223935876565,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":101635.39087210219,"meanTps":101582.39365750182,"stepMs":161.2036895751953,"jitter":0.0007855307490218554,"achievedTflops":130.10417611619724,"nominalPeakTflops":468.0,"mfuNominalPct":27.800037631666076,"configuredPeakTflops":468.0,"mfuConfiguredPct":27.800037631666076,"vramAllocatedGb":22.067508736,"vramAllocatedPct":21.64033916072568,"vramReservedGb":23.290970112,"vramReservedPct":22.840117506502253,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":93023.05117931169,"meanTps":93008.96577309397,"stepMs":352.25677490234375,"jitter":0.00020268504472504385,"achievedTflops":119.07945972017956,"nominalPeakTflops":468.0,"mfuNominalPct":25.444329000038366,"configuredPeakTflops":468.0,"mfuConfiguredPct":25.444329000038366,"vramAllocatedGb":42.006570496,"vramAllocatedPct":41.19343254317644,"vramReservedGb":44.41767936,"vramReservedPct":43.55786861045541,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":91768.78960510445,"meanTps":91766.43554610011,"stepMs":714.142578125,"jitter":0.00014546583944360724,"achievedTflops":117.47387068917176,"nominalPeakTflops":468.0,"mfuNominalPct":25.101254420763198,"configuredPeakTflops":468.0,"mfuConfiguredPct":25.101254420763198,"vramAllocatedGb":81.884694016,"vramAllocatedPct":80.29961930807796,"vramReservedGb":86.50752,"vramReservedPct":84.83295940421556,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":44899.34625217756,"meanTps":44749.4488972239,"stepMs":45.613136291503906,"jitter":0.0010250429824348655,"achievedTflops":57.47596779203103,"nominalPeakTflops":468.0,"mfuNominalPct":12.281189699151929,"configuredPeakTflops":468.0,"mfuConfiguredPct":12.281189699151929,"vramAllocatedGb":4.620813312,"vramAllocatedPct":4.531366384232895,"vramReservedGb":4.657774592,"vramReservedPct":4.567612189982127,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":68113.65751875043,"meanTps":68056.85386414928,"stepMs":60.134782791137695,"jitter":0.002292701379680788,"achievedTflops":87.19277033026431,"nominalPeakTflops":468.0,"mfuNominalPct":18.630933831253056,"configuredPeakTflops":468.0,"mfuConfiguredPct":18.630933831253056,"vramAllocatedGb":7.113212416,"vramAllocatedPct":6.9755191238876115,"vramReservedGb":7.430209536,"vramReservedPct":7.28638000410026,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":90065.20762295913,"meanTps":90062.49218657371,"stepMs":90.9563217163086,"jitter":0.0006626529725687178,"achievedTflops":115.29310345512502,"nominalPeakTflops":468.0,"mfuNominalPct":24.635278516052356,"configuredPeakTflops":468.0,"mfuConfiguredPct":24.635278516052356,"vramAllocatedGb":12.097977856,"vramAllocatedPct":11.863792469500302,"vramReservedGb":12.689866752,"vramReservedPct":12.444223935876565,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":101678.75858452242,"meanTps":101638.10820748478,"stepMs":161.1349334716797,"jitter":0.00024056833910993025,"achievedTflops":130.1596914287873,"nominalPeakTflops":468.0,"mfuNominalPct":27.811899877946008,"configuredPeakTflops":468.0,"mfuConfiguredPct":27.811899877946008,"vramAllocatedGb":22.067508736,"vramAllocatedPct":21.64033916072568,"vramReservedGb":23.290970112,"vramReservedPct":22.840117506502253,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":93027.53623044009,"meanTps":93031.15964550574,"stepMs":352.2397918701172,"jitter":0.0003394945510140685,"achievedTflops":119.08520106556026,"nominalPeakTflops":468.0,"mfuNominalPct":25.44555578323937,"configuredPeakTflops":468.0,"mfuConfiguredPct":25.44555578323937,"vramAllocatedGb":42.006570496,"vramAllocatedPct":41.19343254317644,"vramReservedGb":44.41767936,"vramReservedPct":43.55786861045541,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":91775.34301909972,"meanTps":91770.40924273302,"stepMs":714.0915832519531,"jitter":0.00029329680183892264,"achievedTflops":117.48225976035341,"nominalPeakTflops":468.0,"mfuNominalPct":25.10304695734047,"configuredPeakTflops":468.0,"mfuConfiguredPct":25.10304695734047,"vramAllocatedGb":81.884694016,"vramAllocatedPct":80.29961930807796,"vramReservedGb":86.50752,"vramReservedPct":84.83295940421556,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":59155.1899123312,"meanTps":59130.43863268592,"stepMs":276.9664001464844,"jitter":0.004924828091918637,"achievedTflops":98.0553096366506,"nominalPeakTflops":312.0,"mfuNominalPct":31.427983857900834,"configuredPeakTflops":312.0,"mfuConfiguredPct":31.427983857900834,"vramAllocatedGb":22.074031616,"vramAllocatedPct":25.97732169478885,"vramReservedGb":23.253221376,"vramReservedPct":27.365024325083063,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":61870.27316877146,"meanTps":61845.578383590735,"stepMs":529.6242980957031,"jitter":0.0009057200028879735,"achievedTflops":102.5558163511771,"nominalPeakTflops":312.0,"mfuNominalPct":32.870453958710605,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.870453958710605,"vramAllocatedGb":42.014373376,"vramAllocatedPct":49.44365904605415,"vramReservedGb":44.421873664,"vramReservedPct":52.27687096445792,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":46927.64905482078,"meanTps":47314.994473214036,"stepMs":1396.5327758789062,"jitter":0.030298939376795203,"achievedTflops":77.78700677675145,"nominalPeakTflops":312.0,"mfuNominalPct":24.93173294126649,"configuredPeakTflops":312.0,"mfuConfiguredPct":24.93173294126649,"vramAllocatedGb":81.895056896,"vramAllocatedPct":96.37633374858477,"vramReservedGb":82.715869184,"vramReservedPct":97.34228800774947,"warmupSteps":22,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":59095.03439976352,"meanTps":59134.01718019433,"stepMs":277.2483367919922,"jitter":0.0038115671726194093,"achievedTflops":97.95559619781426,"nominalPeakTflops":312.0,"mfuNominalPct":31.396024422376367,"configuredPeakTflops":312.0,"mfuConfiguredPct":31.396024422376367,"vramAllocatedGb":22.074031616,"vramAllocatedPct":25.97732169478885,"vramReservedGb":23.253221376,"vramReservedPct":27.365024325083063,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":61831.54545950302,"meanTps":61826.41295834824,"stepMs":529.9560241699219,"jitter":0.0009053238186600991,"achievedTflops":102.49162151840817,"nominalPeakTflops":312.0,"mfuNominalPct":32.84987869179749,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.84987869179749,"vramAllocatedGb":42.014373376,"vramAllocatedPct":49.44365904605415,"vramReservedGb":44.421873664,"vramReservedPct":52.27687096445792,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":46538.99076087267,"meanTps":46961.226194046976,"stepMs":1408.195556640625,"jitter":0.03255009502478255,"achievedTflops":77.14276897762657,"nominalPeakTflops":312.0,"mfuNominalPct":24.725246467188004,"configuredPeakTflops":312.0,"mfuConfiguredPct":24.725246467188004,"vramAllocatedGb":81.895056896,"vramAllocatedPct":96.37633374858477,"vramReservedGb":82.715869184,"vramReservedPct":97.34228800774947,"warmupSteps":22,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":179638.4365606945,"meanTps":179657.0913574016,"stepMs":364.8217010498047,"jitter":0.0005137548253363645,"achievedTflops":297.7676607193325,"nominalPeakTflops":2250.0,"mfuNominalPct":13.234118254192557,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.234118254192557,"vramAllocatedGb":81.895056896,"vramAllocatedPct":42.764371098273706,"vramReservedGb":86.553657344,"vramReservedPct":45.197022419462144,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":32,"tokensPerStep":131072,"status":"complete","stable":true,"tps":186058.57047101075,"meanTps":186054.1044863808,"stepMs":704.4663391113281,"jitter":0.00023872965778434915,"achievedTflops":308.4096385308785,"nominalPeakTflops":2250.0,"mfuNominalPct":13.707095045816823,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.707095045816823,"vramAllocatedGb":161.656423936,"vramAllocatedPct":84.41456133790926,"vramReservedGb":170.92837376,"vramReservedPct":89.25623454880456,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.306972672,"vramAllocatedPct":99.37544841405371,"vramReservedGb":190.461247488,"vramReservedPct":99.45600841037827,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 192.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 181.81 MiB is free. Including non-PyTorch memory, this process has 178.16 GiB memory in use. Of the allocated memory 177.24 GiB is allocated by PyTorch, and 107.13 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":179826.38628777146,"meanTps":179828.6232313328,"stepMs":364.4403991699219,"jitter":0.0003731335323455074,"achievedTflops":298.07920512840246,"nominalPeakTflops":2250.0,"mfuNominalPct":13.247964672373444,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.247964672373444,"vramAllocatedGb":81.895056896,"vramAllocatedPct":42.764371098273706,"vramReservedGb":86.553657344,"vramReservedPct":45.197022419462144,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":32,"tokensPerStep":131072,"status":"complete","stable":true,"tps":186082.60874293372,"meanTps":186044.5647136537,"stepMs":704.3753356933594,"jitter":0.00024749642465215176,"achievedTflops":308.4494842350345,"nominalPeakTflops":2250.0,"mfuNominalPct":13.708865966001534,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.708865966001534,"vramAllocatedGb":161.656423936,"vramAllocatedPct":84.41456133790926,"vramReservedGb":170.92837376,"vramReservedPct":89.25623454880456,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.306972672,"vramAllocatedPct":99.37544841405371,"vramReservedGb":190.461247488,"vramReservedPct":99.45600841037827,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 192.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 181.81 MiB is free. Including non-PyTorch memory, this process has 178.16 GiB memory in use. Of the allocated memory 177.24 GiB is allocated by PyTorch, and 107.13 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":181336.78935597718,"meanTps":181316.96279033928,"stepMs":361.4048767089844,"jitter":0.00033689436745488603,"achievedTflops":300.5828407476701,"nominalPeakTflops":2250.0,"mfuNominalPct":13.359237366563114,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.359237366563114,"vramAllocatedGb":81.895056896,"vramAllocatedPct":28.49230914399451,"vramReservedGb":86.553657344,"vramReservedPct":30.11309419713061,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":32,"tokensPerStep":131072,"status":"complete","stable":true,"tps":189116.01508254636,"meanTps":189106.59420639733,"stepMs":693.0772094726562,"jitter":7.435602534376446e-05,"achievedTflops":313.47764149943185,"nominalPeakTflops":2250.0,"mfuNominalPct":13.93233962219697,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.93233962219697,"vramAllocatedGb":161.656423936,"vramAllocatedPct":56.24228104202117,"vramReservedGb":170.92837376,"vramReservedPct":59.46810773737959,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":270.859795968,"vramAllocatedPct":94.23549276240232,"vramReservedGb":272.992567296,"vramReservedPct":94.97751044105163,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 31.24 GiB. GPU 0 has a total capacity of 267.69 GiB of which 14.43 GiB is free. Including non-PyTorch memory, this process has 253.25 GiB memory in use. Of the allocated memory 252.26 GiB is allocated by PyTorch, and 173.97 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":181390.96533501457,"meanTps":181374.8041189987,"stepMs":361.29693603515625,"jitter":0.00015910522850874458,"achievedTflops":300.6726425454034,"nominalPeakTflops":2250.0,"mfuNominalPct":13.363228557573484,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.363228557573484,"vramAllocatedGb":81.895056896,"vramAllocatedPct":28.49230914399451,"vramReservedGb":86.553657344,"vramReservedPct":30.11309419713061,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":32,"tokensPerStep":131072,"status":"complete","stable":true,"tps":189142.33256064027,"meanTps":189138.86423092356,"stepMs":692.9807739257812,"jitter":0.00019104127015834437,"achievedTflops":313.52126520290034,"nominalPeakTflops":2250.0,"mfuNominalPct":13.934278453462236,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.934278453462236,"vramAllocatedGb":161.656423936,"vramAllocatedPct":56.24228104202117,"vramReservedGb":170.92837376,"vramReservedPct":59.46810773737959,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":270.859795968,"vramAllocatedPct":94.23549276240232,"vramReservedGb":272.992567296,"vramReservedPct":94.97751044105163,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 31.24 GiB. GPU 0 has a total capacity of 267.69 GiB of which 14.43 GiB is free. Including non-PyTorch memory, this process has 253.25 GiB memory in use. Of the allocated memory 252.26 GiB is allocated by PyTorch, and 173.97 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":50532.8865352869,"meanTps":50520.27594900044,"stepMs":81.05612564086914,"jitter":0.001373376225393591,"achievedTflops":83.76302811967462,"nominalPeakTflops":165.2,"mfuNominalPct":50.704012178979795,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":40.75729208645266,"vramAllocatedGb":7.118742528,"vramAllocatedPct":28.180709613104582,"vramReservedGb":7.178551296,"vramReservedPct":28.417472428545114,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":52880.395478929386,"meanTps":52762.91004973992,"stepMs":154.91563415527344,"jitter":0.0008435553684980258,"achievedTflops":87.65424572348601,"nominalPeakTflops":165.2,"mfuNominalPct":53.059470776928585,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":42.65067507427752,"vramAllocatedGb":12.103860736,"vramAllocatedPct":47.9151174883838,"vramReservedGb":12.69825536,"vramReservedPct":50.26812607503379,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":50380.921328260934,"meanTps":50336.8574714916,"stepMs":325.2024688720703,"jitter":0.0032227336119527022,"achievedTflops":83.5111314483764,"nominalPeakTflops":165.2,"mfuNominalPct":50.55153235373874,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":40.63472457142636,"vramAllocatedGb":22.074031616,"vramAllocatedPct":87.38367380393979,"vramReservedGb":23.297261568,"vramReservedPct":92.22603015153598,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":50439.51289833309,"meanTps":50397.51936897611,"stepMs":81.2061767578125,"jitter":0.0028329130721046957,"achievedTflops":83.60825250494028,"nominalPeakTflops":165.2,"mfuNominalPct":50.61032233955223,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":40.68198159351567,"vramAllocatedGb":7.118742528,"vramAllocatedPct":28.180709613104582,"vramReservedGb":7.178551296,"vramReservedPct":28.417472428545114,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":52849.677490231785,"meanTps":52843.83621208403,"stepMs":155.00567626953125,"jitter":0.0012296801028345657,"achievedTflops":87.60332775842458,"nominalPeakTflops":165.2,"mfuNominalPct":53.028648764179536,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":42.625899485082115,"vramAllocatedGb":12.103860736,"vramAllocatedPct":47.9151174883838,"vramReservedGb":12.69825536,"vramReservedPct":50.26812607503379,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":50463.967677247725,"meanTps":50470.02150022835,"stepMs":324.66729736328125,"jitter":0.0005345817317119069,"achievedTflops":83.64878860873986,"nominalPeakTflops":165.2,"mfuNominalPct":50.63485993265125,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":40.70170559179623,"vramAllocatedGb":22.074031616,"vramAllocatedPct":87.38367380393979,"vramReservedGb":23.297261568,"vramReservedPct":92.22603015153598,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":72975.02525855614,"meanTps":72876.46429656754,"stepMs":112.25758361816406,"jitter":0.00047353276435374456,"achievedTflops":120.9629908732404,"nominalPeakTflops":209.5,"mfuNominalPct":57.73889779152286,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":44.97654594957784,"vramAllocatedGb":12.103860736,"vramAllocatedPct":35.94972181048612,"vramReservedGb":12.696158208,"vramReservedPct":37.70890673601352,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":79633.66706428798,"meanTps":79604.63725026742,"stepMs":205.74212646484375,"jitter":0.0005123972115285063,"achievedTflops":132.00031802895086,"nominalPeakTflops":209.5,"mfuNominalPct":63.007311708329766,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":49.08045284205493,"vramAllocatedGb":22.074031616,"vramAllocatedPct":65.56216344020197,"vramReservedGb":23.253221376,"vramReservedPct":69.06447933414567,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":31.631572992,"vramAllocatedPct":93.94905264469213,"vramReservedGb":31.736201216,"vramReservedPct":94.25980932211638,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 1.30 GiB is free. Including non-PyTorch memory, this process has 30.04 GiB memory in use. Of the allocated memory 29.37 GiB is allocated by PyTorch, and 95.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":72950.40408207414,"meanTps":72950.53963649047,"stepMs":112.29547119140625,"jitter":0.0004078995545286597,"achievedTflops":120.92217894977028,"nominalPeakTflops":209.5,"mfuNominalPct":57.71941715979488,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":44.961371230947016,"vramAllocatedGb":12.103860736,"vramAllocatedPct":35.94972181048612,"vramReservedGb":12.696158208,"vramReservedPct":37.70890673601352,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":79365.51043203365,"meanTps":79364.57799570145,"stepMs":206.4372787475586,"jitter":0.00046053608179451457,"achievedTflops":131.5558231055842,"nominalPeakTflops":209.5,"mfuNominalPct":62.79514229383494,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":48.91518042111997,"vramAllocatedGb":22.074031616,"vramAllocatedPct":65.56216344020197,"vramReservedGb":23.253221376,"vramReservedPct":69.06447933414567,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":31.631572992,"vramAllocatedPct":93.94905264469213,"vramReservedGb":31.736201216,"vramReservedPct":94.25980932211638,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 1.30 GiB is free. Including non-PyTorch memory, this process has 30.04 GiB memory in use. Of the allocated memory 29.37 GiB is allocated by PyTorch, and 95.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":116971.68582783685,"meanTps":116954.35130983415,"stepMs":140.06808471679688,"jitter":0.0006768969166301178,"achievedTflops":193.8916076436878,"nominalPeakTflops":989.5,"mfuNominalPct":19.594907290923476,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.594907290923476,"vramAllocatedGb":22.12410112,"vramAllocatedPct":26.02299857141646,"vramReservedGb":23.295164416,"vramReservedPct":27.400436611206356,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":126061.40537436002,"meanTps":126055.214799839,"stepMs":259.9368133544922,"jitter":0.0002595278170477292,"achievedTflops":208.9586755707041,"nominalPeakTflops":989.5,"mfuNominalPct":21.11760238208227,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.11760238208227,"vramAllocatedGb":42.06444288,"vramAllocatedPct":49.477397117124966,"vramReservedGb":44.442845184,"vramReservedPct":52.274941723504234,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":114216.30440453866,"meanTps":113908.27244666588,"stepMs":573.7884826660156,"jitter":0.007153610265259879,"achievedTflops":189.32430291473688,"nominalPeakTflops":989.5,"mfuNominalPct":19.133330259195237,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.133330259195237,"vramAllocatedGb":81.9451264,"vramAllocatedPct":96.38619420854198,"vramReservedGb":82.673926144,"vramReservedPct":97.24342924801736,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":116886.25109491547,"meanTps":116828.43595862474,"stepMs":140.17046356201172,"jitter":0.0006063574956879667,"achievedTflops":193.74999151156572,"nominalPeakTflops":989.5,"mfuNominalPct":19.580595402886882,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.580595402886882,"vramAllocatedGb":22.12410112,"vramAllocatedPct":26.02299857141646,"vramReservedGb":23.295164416,"vramReservedPct":27.400436611206356,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":125921.89886157528,"meanTps":125888.66558469785,"stepMs":260.22479248046875,"jitter":0.0006753991593337246,"achievedTflops":208.72743036081278,"nominalPeakTflops":989.5,"mfuNominalPct":21.094232477090728,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.094232477090728,"vramAllocatedGb":42.06444288,"vramAllocatedPct":49.477397117124966,"vramReservedGb":44.442845184,"vramReservedPct":52.274941723504234,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":113824.85309194117,"meanTps":113193.85067075207,"stepMs":575.7617797851562,"jitter":0.007192824034394819,"achievedTflops":188.67543542362907,"nominalPeakTflops":989.5,"mfuNominalPct":19.06775496954311,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.06775496954311,"vramAllocatedGb":81.9451264,"vramAllocatedPct":96.38619420854198,"vramReservedGb":82.673926144,"vramReservedPct":97.24342924801736,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":139651.58877571154,"meanTps":139660.03334879372,"stepMs":234.64108276367188,"jitter":0.0005749165628240544,"achievedTflops":231.48568703687144,"nominalPeakTflops":989.5,"mfuNominalPct":23.394207886495344,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.394207886495344,"vramAllocatedGb":42.06444288,"vramAllocatedPct":28.02040375845662,"vramReservedGb":44.373639168,"vramReservedPct":29.558629583338607,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":145798.31997641953,"meanTps":145772.42035333952,"stepMs":449.4976348876953,"jitter":0.0006658501445960332,"achievedTflops":241.6744740567749,"nominalPeakTflops":989.5,"mfuNominalPct":24.42389833822889,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.42389833822889,"vramAllocatedGb":81.9451264,"vramAllocatedPct":54.58613904185299,"vramReservedGb":86.69626368,"vramReservedPct":57.751015973118676,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.546811904,"vramAllocatedPct":90.95798112422571,"vramReservedGb":137.325707264,"vramReservedPct":91.47682699447888,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.83 GiB is free. Including non-PyTorch memory, this process has 127.97 GiB memory in use. Of the allocated memory 127.17 GiB is allocated by PyTorch, and 82.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":139196.00346595826,"meanTps":139180.13842368955,"stepMs":235.4090576171875,"jitter":0.0004840048735500819,"achievedTflops":230.7305113932809,"nominalPeakTflops":989.5,"mfuNominalPct":23.31788897355037,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.31788897355037,"vramAllocatedGb":42.06444288,"vramAllocatedPct":28.02040375845662,"vramReservedGb":44.373639168,"vramReservedPct":29.558629583338607,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":145752.08328653054,"meanTps":145755.18411105312,"stepMs":449.6402282714844,"jitter":0.0009534284782124596,"achievedTflops":241.59783238001998,"nominalPeakTflops":989.5,"mfuNominalPct":24.416152842851943,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.416152842851943,"vramAllocatedGb":81.9451264,"vramAllocatedPct":54.58613904185299,"vramReservedGb":86.69626368,"vramReservedPct":57.751015973118676,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.546811904,"vramAllocatedPct":90.95798112422571,"vramReservedGb":137.325707264,"vramReservedPct":91.47682699447888,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.83 GiB is free. Including non-PyTorch memory, this process has 127.97 GiB memory in use. Of the allocated memory 127.17 GiB is allocated by PyTorch, and 82.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":49488.1822473166,"meanTps":49454.281219601646,"stepMs":82.7672348022461,"jitter":0.0028319611609640325,"achievedTflops":82.03133217570958,"nominalPeakTflops":364.2,"mfuNominalPct":22.523704606180555,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":18.171297918995904,"vramAllocatedGb":7.118742528,"vramAllocatedPct":13.99210770636813,"vramReservedGb":7.178551296,"vramReservedPct":14.10966368206884,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":51766.222681448555,"meanTps":51821.827733367274,"stepMs":158.24990844726562,"jitter":0.0034480845074170772,"achievedTflops":85.80739916940298,"nominalPeakTflops":364.2,"mfuNominalPct":23.560515971829485,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":19.007759262297217,"vramAllocatedGb":12.103860736,"vramAllocatedPct":23.79051109305582,"vramReservedGb":12.69825536,"vramReservedPct":24.95881203474345,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":50408.09029367754,"meanTps":50453.07198936415,"stepMs":325.0271911621094,"jitter":0.006436066628289719,"achievedTflops":83.55616657243536,"nominalPeakTflops":364.2,"mfuNominalPct":22.942385110498453,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":18.509073978034984,"vramAllocatedGb":22.074031616,"vramAllocatedPct":43.38718905340459,"vramReservedGb":23.297261568,"vramReservedPct":45.79148520131544,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":32321.564311206643,"meanTps":32320.301001443306,"stepMs":506.9061584472656,"jitter":0.000945642448020659,"achievedTflops":53.57604296720593,"nominalPeakTflops":154.8,"mfuNominalPct":34.60984687804,"configuredPeakTflops":180.6,"mfuConfiguredPct":29.66558303832001,"vramAllocatedGb":22.074031616,"vramAllocatedPct":43.25363703680161,"vramReservedGb":23.253221376,"vramReservedPct":45.56423651241274,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":33453.91969826656,"meanTps":33452.252304412745,"stepMs":979.49658203125,"jitter":0.0013111731116360892,"achievedTflops":55.453028879370905,"nominalPeakTflops":154.8,"mfuNominalPct":35.822370077112986,"configuredPeakTflops":180.6,"mfuConfiguredPct":30.704888637525418,"vramAllocatedGb":42.014373376,"vramAllocatedPct":82.3263501632816,"vramReservedGb":44.421873664,"vramReservedPct":87.04380030717232,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.407634944,"vramAllocatedPct":98.77278349873382,"vramReservedGb":50.63573504,"vramReservedPct":99.21974121502576,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 164.25 MiB is free. Process 1415479 has 47.36 GiB memory in use. Of the allocated memory 46.95 GiB is allocated by PyTorch, and 97.53 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":32386.467683096234,"meanTps":32392.135471720678,"stepMs":505.8903045654297,"jitter":0.0009392684152667185,"achievedTflops":53.68362643091432,"nominalPeakTflops":154.8,"mfuNominalPct":34.67934523960873,"configuredPeakTflops":180.6,"mfuConfiguredPct":29.725153062521773,"vramAllocatedGb":22.074031616,"vramAllocatedPct":43.25363703680161,"vramReservedGb":23.253221376,"vramReservedPct":45.56423651241274,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":33505.470037815365,"meanTps":33501.2748989689,"stepMs":977.9895629882812,"jitter":0.0006325165456653624,"achievedTflops":55.5384784318755,"nominalPeakTflops":154.8,"mfuNominalPct":35.8775700464312,"configuredPeakTflops":180.6,"mfuConfiguredPct":30.752202896941032,"vramAllocatedGb":42.014373376,"vramAllocatedPct":82.3263501632816,"vramReservedGb":44.421873664,"vramReservedPct":87.04380030717232,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.407634944,"vramAllocatedPct":98.77278349873382,"vramReservedGb":50.63573504,"vramReservedPct":99.21974121502576,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 164.25 MiB is free. Process 1449495 has 47.36 GiB memory in use. Of the allocated memory 46.95 GiB is allocated by PyTorch, and 97.53 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":85300.39385050777,"meanTps":85275.7011811686,"stepMs":96.03707122802734,"jitter":0.0008227590333507019,"achievedTflops":141.3934524347835,"nominalPeakTflops":468.0,"mfuNominalPct":30.212276161278528,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.212276161278528,"vramAllocatedGb":12.103860736,"vramAllocatedPct":11.869561472243877,"vramReservedGb":12.696158208,"vramReservedPct":12.450393605651417,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":94849.04788276984,"meanTps":94836.43196456095,"stepMs":172.73763275146484,"jitter":0.0002723428803068336,"achievedTflops":157.22124758064163,"nominalPeakTflops":468.0,"mfuNominalPct":33.59428367107727,"configuredPeakTflops":468.0,"mfuConfiguredPct":33.59428367107727,"vramAllocatedGb":22.074031616,"vramAllocatedPct":21.646735774733788,"vramReservedGb":23.253221376,"vramReservedPct":22.803099487853142,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":87647.19935781765,"meanTps":87651.77558257875,"stepMs":373.86248779296875,"jitter":0.00022395009895740054,"achievedTflops":145.28350402649176,"nominalPeakTflops":468.0,"mfuNominalPct":31.04348376634439,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.04348376634439,"vramAllocatedGb":42.014373376,"vramAllocatedPct":41.20108437971361,"vramReservedGb":44.33379328,"vramReservedPct":43.47560634679071,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":85174.01761384333,"meanTps":85163.61506487335,"stepMs":96.1795654296875,"jitter":0.00041356319400992253,"achievedTflops":141.1839718966395,"nominalPeakTflops":468.0,"mfuNominalPct":30.167515362529805,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.167515362529805,"vramAllocatedGb":12.122853888,"vramAllocatedPct":11.88818696621913,"vramReservedGb":13.688111104,"vramReservedPct":13.423144873486423,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":94679.97431210232,"meanTps":94687.5534082102,"stepMs":173.0460968017578,"jitter":0.00018831959864379328,"achievedTflops":156.94099218212546,"nominalPeakTflops":468.0,"mfuNominalPct":33.53440003891569,"configuredPeakTflops":468.0,"mfuConfiguredPct":33.53440003891569,"vramAllocatedGb":22.09064704,"vramAllocatedPct":21.663029567339052,"vramReservedGb":25.423773696,"vramReservedPct":24.931635560177096,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":87605.06917997563,"meanTps":87605.78966126648,"stepMs":374.0422821044922,"jitter":0.00022686653148799003,"achievedTflops":145.213669280978,"nominalPeakTflops":468.0,"mfuNominalPct":31.028561812174786,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.028561812174786,"vramAllocatedGb":42.03489024,"vramAllocatedPct":41.22120409058746,"vramReservedGb":44.434456576,"vramReservedPct":43.57432106318834,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":50107.37054116839,"meanTps":49998.73897447235,"stepMs":326.97784423828125,"jitter":0.003433137923600086,"achievedTflops":120.88749324858469,"nominalPeakTflops":312.0,"mfuNominalPct":38.74599142582843,"configuredPeakTflops":312.0,"mfuConfiguredPct":38.74599142582843,"vramAllocatedGb":22.085157376,"vramAllocatedPct":25.99041479221875,"vramReservedGb":23.261609984,"vramReservedPct":27.374896267480278,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":52195.683069385224,"meanTps":52215.001241182516,"stepMs":627.7913818359375,"jitter":0.002220426048350074,"achievedTflops":125.92569150024374,"nominalPeakTflops":312.0,"mfuNominalPct":40.36079855777043,"configuredPeakTflops":312.0,"mfuConfiguredPct":40.36079855777043,"vramAllocatedGb":42.026139136,"vramAllocatedPct":49.457505312917434,"vramReservedGb":44.451233792,"vramReservedPct":52.31142276284818,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":41738.34764329793,"meanTps":42657.4084560376,"stepMs":1570.1627807617188,"jitter":0.04007810502921473,"achievedTflops":100.69664730841824,"nominalPeakTflops":312.0,"mfuNominalPct":32.274566445005846,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.274566445005846,"vramAllocatedGb":81.908102656,"vramAllocatedPct":96.39168635431481,"vramReservedGb":84.213235712,"vramReservedPct":99.10442972565255,"warmupSteps":45,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":50137.30015668639,"meanTps":50104.813254567605,"stepMs":326.78265380859375,"jitter":0.00470393972215576,"achievedTflops":120.95970051379913,"nominalPeakTflops":312.0,"mfuNominalPct":38.76913478006383,"configuredPeakTflops":312.0,"mfuConfiguredPct":38.76913478006383,"vramAllocatedGb":22.085157376,"vramAllocatedPct":25.99041479221875,"vramReservedGb":23.261609984,"vramReservedPct":27.374896267480278,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":52171.18693257177,"meanTps":52169.691381919205,"stepMs":628.0861511230469,"jitter":0.0033064203626944236,"achievedTflops":125.86659287779219,"nominalPeakTflops":312.0,"mfuNominalPct":40.341856691600064,"configuredPeakTflops":312.0,"mfuConfiguredPct":40.341856691600064,"vramAllocatedGb":42.026139136,"vramAllocatedPct":49.457505312917434,"vramReservedGb":44.451233792,"vramReservedPct":52.31142276284818,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":41581.07833733437,"meanTps":41050.371921495425,"stepMs":1576.1015014648438,"jitter":0.04288656762749669,"achievedTflops":100.31722424235934,"nominalPeakTflops":312.0,"mfuNominalPct":32.15295648793569,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.15295648793569,"vramAllocatedGb":81.908102656,"vramAllocatedPct":96.39168635431481,"vramReservedGb":82.178998272,"vramReservedPct":96.71048369432765,"warmupSteps":16,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":143302.82720446886,"meanTps":143269.04496004,"stepMs":457.3252410888672,"jitter":0.00032079594766547574,"achievedTflops":345.72797113650705,"nominalPeakTflops":2250.0,"mfuNominalPct":15.36568760606698,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.36568760606698,"vramAllocatedGb":81.908102656,"vramAllocatedPct":42.771183398588825,"vramReservedGb":86.564143104,"vramReservedPct":45.20249792615184,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":16,"tokensPerStep":131072,"status":"complete","stable":true,"tps":147594.306039794,"meanTps":147581.27620476703,"stepMs":888.0559387207031,"jitter":0.00023181692173528101,"achievedTflops":356.0814603164187,"nominalPeakTflops":2250.0,"mfuNominalPct":15.825842680729721,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.825842680729721,"vramAllocatedGb":161.672029696,"vramAllocatedPct":84.4227104318498,"vramReservedGb":170.93885952,"vramReservedPct":89.26171005549426,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.3174912,"vramAllocatedPct":99.38094103170181,"vramReservedGb":190.469636096,"vramReservedPct":99.46038881573004,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 192.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 173.81 MiB is free. Including non-PyTorch memory, this process has 178.17 GiB memory in use. Of the allocated memory 177.25 GiB is allocated by PyTorch, and 105.10 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":142940.7117350126,"meanTps":143008.92217649915,"stepMs":458.4837951660156,"jitter":0.0014226887808558761,"achievedTflops":344.85434254861025,"nominalPeakTflops":2250.0,"mfuNominalPct":15.32685966882712,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.32685966882712,"vramAllocatedGb":81.908102656,"vramAllocatedPct":42.771183398588825,"vramReservedGb":86.564143104,"vramReservedPct":45.20249792615184,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":16,"tokensPerStep":131072,"status":"complete","stable":true,"tps":147603.91303530495,"meanTps":147585.6597987257,"stepMs":887.9981384277344,"jitter":0.00011540648451093914,"achievedTflops":356.104637856817,"nominalPeakTflops":2250.0,"mfuNominalPct":15.82687279363631,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.82687279363631,"vramAllocatedGb":161.672029696,"vramAllocatedPct":84.4227104318498,"vramReservedGb":170.93885952,"vramReservedPct":89.26171005549426,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.3174912,"vramAllocatedPct":99.38094103170181,"vramReservedGb":190.469636096,"vramReservedPct":99.46038881573004,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 192.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 173.81 MiB is free. Including non-PyTorch memory, this process has 178.17 GiB memory in use. Of the allocated memory 177.25 GiB is allocated by PyTorch, and 105.10 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":144606.751520108,"meanTps":144603.82351446705,"stepMs":453.20152282714844,"jitter":0.00039993889771107146,"achievedTflops":348.8737786335096,"nominalPeakTflops":2250.0,"mfuNominalPct":15.505501272600425,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.505501272600425,"vramAllocatedGb":81.908102656,"vramAllocatedPct":28.496847926199774,"vramReservedGb":86.564143104,"vramReservedPct":30.116742323487117,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":16,"tokensPerStep":131072,"status":"complete","stable":true,"tps":149996.88465099354,"meanTps":149996.11355189688,"stepMs":873.8314819335938,"jitter":0.0001354515918475112,"achievedTflops":361.8778472052887,"nominalPeakTflops":2250.0,"mfuNominalPct":16.08345987579061,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.08345987579061,"vramAllocatedGb":161.672029696,"vramAllocatedPct":56.247710480075185,"vramReservedGb":170.93885952,"vramReservedPct":59.4717558637361,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":270.87437824,"vramAllocatedPct":94.240566116248,"vramReservedGb":273.003053056,"vramReservedPct":94.98115856740814,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 31.25 GiB. GPU 0 has a total capacity of 267.69 GiB of which 14.42 GiB is free. Including non-PyTorch memory, this process has 253.26 GiB memory in use. Of the allocated memory 252.27 GiB is allocated by PyTorch, and 170.06 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":144614.2497805096,"meanTps":144625.28970610737,"stepMs":453.1780242919922,"jitter":0.00019783325462823835,"achievedTflops":348.8918687047682,"nominalPeakTflops":2250.0,"mfuNominalPct":15.506305275767476,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.506305275767476,"vramAllocatedGb":81.908102656,"vramAllocatedPct":28.496847926199774,"vramReservedGb":86.564143104,"vramReservedPct":30.116742323487117,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":16,"tokensPerStep":131072,"status":"complete","stable":true,"tps":150001.26414135026,"meanTps":149982.8023415724,"stepMs":873.8059692382812,"jitter":0.00010288615107236527,"achievedTflops":361.8884130283445,"nominalPeakTflops":2250.0,"mfuNominalPct":16.08392946792642,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.08392946792642,"vramAllocatedGb":161.672029696,"vramAllocatedPct":56.247710480075185,"vramReservedGb":170.93885952,"vramReservedPct":59.4717558637361,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":270.87437824,"vramAllocatedPct":94.240566116248,"vramReservedGb":273.003053056,"vramReservedPct":94.98115856740814,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 31.25 GiB. GPU 0 has a total capacity of 267.69 GiB of which 14.42 GiB is free. Including non-PyTorch memory, this process has 253.26 GiB memory in use. Of the allocated memory 252.27 GiB is allocated by PyTorch, and 170.06 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":45642.51759880485,"meanTps":45621.10528490167,"stepMs":179.48177337646484,"jitter":0.0012147533468069229,"achievedTflops":110.11572705737257,"nominalPeakTflops":165.2,"mfuNominalPct":66.65600911463231,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":53.57994991032645,"vramAllocatedGb":12.11460096,"vramAllocatedPct":47.95763442624652,"vramReservedGb":12.180258816,"vramReservedPct":48.21755181565587,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":43875.12751465645,"meanTps":43758.795493172045,"stepMs":373.4234161376953,"jitter":0.0014171911720160405,"achievedTflops":105.85177637391841,"nominalPeakTflops":165.2,"mfuNominalPct":64.07492516581019,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":51.505203004096224,"vramAllocatedGb":22.085157376,"vramAllocatedPct":87.42771694927787,"vramReservedGb":23.305650176,"vramReservedPct":92.2592378318498,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":45616.00896837745,"meanTps":45623.298437182246,"stepMs":179.58607482910156,"jitter":0.0013215013008786463,"achievedTflops":110.0517731550382,"nominalPeakTflops":165.2,"mfuNominalPct":66.61729609869141,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":53.54883131378066,"vramAllocatedGb":12.11460096,"vramAllocatedPct":47.95763442624652,"vramReservedGb":12.180258816,"vramReservedPct":48.21755181565587,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":43888.87917433425,"meanTps":43884.12537640708,"stepMs":373.30641174316406,"jitter":0.0003403342954012402,"achievedTflops":105.88495320296578,"nominalPeakTflops":165.2,"mfuNominalPct":64.09500799211004,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":51.5213461371985,"vramAllocatedGb":22.085157376,"vramAllocatedPct":87.42771694927787,"vramReservedGb":23.305650176,"vramReservedPct":92.2592378318498,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":62615.36251844725,"meanTps":62605.33111586428,"stepMs":130.83051300048828,"jitter":0.000821350697859725,"achievedTflops":151.06388804591967,"nominalPeakTflops":209.5,"mfuNominalPct":72.10686780234829,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":56.16868310687777,"vramAllocatedGb":12.11460096,"vramAllocatedPct":35.98162138975291,"vramReservedGb":12.180258816,"vramReservedPct":36.17663203217154,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":68188.03771717857,"meanTps":68169.22046012501,"stepMs":240.27674865722656,"jitter":0.0006311519557303695,"achievedTflops":164.50835196784308,"nominalPeakTflops":209.5,"mfuNominalPct":78.52427301567688,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":61.16761331674248,"vramAllocatedGb":22.085157376,"vramAllocatedPct":65.59520810137694,"vramReservedGb":23.263707136,"vramReservedPct":69.09562312893921,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":31.642058752,"vramAllocatedPct":93.98019643948567,"vramReservedGb":31.75088128,"vramReservedPct":94.30341063482733,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 1.29 GiB is free. Including non-PyTorch memory, this process has 30.06 GiB memory in use. Of the allocated memory 29.38 GiB is allocated by PyTorch, and 99.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":62373.70203621844,"meanTps":62358.14062527232,"stepMs":131.33740234375,"jitter":0.0003947937385446447,"achievedTflops":150.48086543670328,"nominalPeakTflops":209.5,"mfuNominalPct":71.82857538744787,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":55.951903222520066,"vramAllocatedGb":12.11460096,"vramAllocatedPct":35.98162138975291,"vramReservedGb":12.180258816,"vramReservedPct":36.17663203217154,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":67975.46407337215,"meanTps":67976.42846295486,"stepMs":241.02814483642578,"jitter":0.00037919069552992793,"achievedTflops":163.99550336587222,"nominalPeakTflops":209.5,"mfuNominalPct":78.2794765469557,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":60.97692558204608,"vramAllocatedGb":22.085157376,"vramAllocatedPct":65.59520810137694,"vramReservedGb":23.263707136,"vramReservedPct":69.09562312893921,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":31.642058752,"vramAllocatedPct":93.98019643948567,"vramReservedGb":31.75088128,"vramReservedPct":94.30341063482733,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 1.29 GiB is free. Including non-PyTorch memory, this process has 30.06 GiB memory in use. Of the allocated memory 29.38 GiB is allocated by PyTorch, and 99.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":98518.14949806053,"meanTps":98511.91958577062,"stepMs":166.30438232421875,"jitter":0.0007919089902819763,"achievedTflops":237.68184208598356,"nominalPeakTflops":989.5,"mfuNominalPct":24.020398391711325,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.020398391711325,"vramAllocatedGb":22.13522688,"vramAllocatedPct":26.03608500756207,"vramReservedGb":23.303553024,"vramReservedPct":27.410303531123965,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":105585.75504321478,"meanTps":105574.98563591606,"stepMs":310.3448944091797,"jitter":0.0005904518381968461,"achievedTflops":254.73292875040005,"nominalPeakTflops":989.5,"mfuNominalPct":25.743600682203137,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.743600682203137,"vramAllocatedGb":42.07620864,"vramAllocatedPct":49.491236339519176,"vramReservedGb":44.472205312,"vramReservedPct":52.30947594321587,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":97540.1325856426,"meanTps":97386.00141415981,"stepMs":671.8875427246094,"jitter":0.005273448338151027,"achievedTflops":235.32230871554285,"nominalPeakTflops":989.5,"mfuNominalPct":23.781941254728938,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.781941254728938,"vramAllocatedGb":81.95817216,"vramAllocatedPct":96.40153900343338,"vramReservedGb":82.682314752,"vramReservedPct":97.25329616793498,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":98384.38317561551,"meanTps":98365.30315530949,"stepMs":166.5304946899414,"jitter":0.0014535637064920673,"achievedTflops":237.35912159143726,"nominalPeakTflops":989.5,"mfuNominalPct":23.987783889988606,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.987783889988606,"vramAllocatedGb":22.13522688,"vramAllocatedPct":26.03608500756207,"vramReservedGb":23.303553024,"vramReservedPct":27.410303531123965,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":105466.05364434767,"meanTps":105433.76722314206,"stepMs":310.69712829589844,"jitter":0.0007907453582182402,"achievedTflops":254.44414085570295,"nominalPeakTflops":989.5,"mfuNominalPct":25.714415447771902,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.714415447771902,"vramAllocatedGb":42.07620864,"vramAllocatedPct":49.491236339519176,"vramReservedGb":44.472205312,"vramReservedPct":52.30947594321587,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":96908.82405447405,"meanTps":96452.49764330732,"stepMs":676.2645263671875,"jitter":0.010447590207797565,"achievedTflops":233.79923326825494,"nominalPeakTflops":989.5,"mfuNominalPct":23.628017510687716,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.628017510687716,"vramAllocatedGb":81.95817216,"vramAllocatedPct":96.40153900343338,"vramReservedGb":82.682314752,"vramReservedPct":97.25329616793498,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":114884.54713236625,"meanTps":114881.32732719019,"stepMs":285.22547912597656,"jitter":0.00023358289995627666,"achievedTflops":277.1669071004258,"nominalPeakTflops":989.5,"mfuNominalPct":28.010804153655968,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.010804153655968,"vramAllocatedGb":42.07620864,"vramAllocatedPct":28.02824128875901,"vramReservedGb":44.384124928,"vramReservedPct":29.56561446674126,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":119510.18471044896,"meanTps":119347.2467847704,"stepMs":548.3716735839844,"jitter":0.00131386643128407,"achievedTflops":288.3265773335994,"nominalPeakTflops":989.5,"mfuNominalPct":29.138613171662396,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.138613171662396,"vramAllocatedGb":81.95817216,"vramAllocatedPct":54.594829219055114,"vramReservedGb":86.708846592,"vramReservedPct":57.75939783320186,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.559345664,"vramAllocatedPct":90.96633024266795,"vramReservedGb":137.334095872,"vramReservedPct":91.48241490120101,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.82 GiB is free. Including non-PyTorch memory, this process has 127.98 GiB memory in use. Of the allocated memory 127.18 GiB is allocated by PyTorch, and 78.86 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":114674.84150988533,"meanTps":114669.99366498861,"stepMs":285.7470703125,"jitter":0.000256986786794484,"achievedTflops":276.6609777980486,"nominalPeakTflops":989.5,"mfuNominalPct":27.959674360591066,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.959674360591066,"vramAllocatedGb":42.07620864,"vramAllocatedPct":28.02824128875901,"vramReservedGb":44.384124928,"vramReservedPct":29.56561446674126,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":119505.38295854176,"meanTps":119501.55646467333,"stepMs":548.3937072753906,"jitter":0.0015639695677711978,"achievedTflops":288.31499277537984,"nominalPeakTflops":989.5,"mfuNominalPct":29.137442422979266,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.137442422979266,"vramAllocatedGb":81.95817216,"vramAllocatedPct":54.594829219055114,"vramReservedGb":86.708846592,"vramReservedPct":57.75939783320186,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.559345664,"vramAllocatedPct":90.96633024266795,"vramReservedGb":137.334095872,"vramReservedPct":91.48241490120101,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.82 GiB is free. Including non-PyTorch memory, this process has 127.98 GiB memory in use. Of the allocated memory 127.18 GiB is allocated by PyTorch, and 78.86 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":45711.114286020806,"meanTps":45792.81787097868,"stepMs":179.21243286132812,"jitter":0.00288848363191202,"achievedTflops":110.28122130448898,"nominalPeakTflops":364.2,"mfuNominalPct":30.2804012368174,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":24.42911597366428,"vramAllocatedGb":12.11460096,"vramAllocatedPct":23.81162133414228,"vramReservedGb":12.180258816,"vramReservedPct":23.940673872467375,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":44034.46373550266,"meanTps":43979.6566396969,"stepMs":372.07220458984375,"jitter":0.005487914735013651,"achievedTflops":106.23618601493072,"nominalPeakTflops":364.2,"mfuNominalPct":29.169738060112774,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":23.53307370067069,"vramAllocatedGb":22.085157376,"vramAllocatedPct":43.40905707737412,"vramReservedGb":23.305650176,"vramReservedPct":45.80797326872072,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":28397.00518869798,"meanTps":28393.96332682608,"stepMs":576.9622497558594,"jitter":0.0010033050742522215,"achievedTflops":68.50973691002835,"nominalPeakTflops":154.8,"mfuNominalPct":44.2569359883904,"configuredPeakTflops":180.6,"mfuConfiguredPct":37.934516561477494,"vramAllocatedGb":22.085157376,"vramAllocatedPct":43.27543774784389,"vramReservedGb":23.261609984,"vramReservedPct":45.58067382717191,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":29320.620525448092,"meanTps":29315.336822038324,"stepMs":1117.5752563476562,"jitter":0.000818612617681374,"achievedTflops":70.73802272067438,"nominalPeakTflops":154.8,"mfuNominalPct":45.6963971063788,"configuredPeakTflops":180.6,"mfuConfiguredPct":39.16834037689611,"vramAllocatedGb":42.026139136,"vramAllocatedPct":82.34940494191721,"vramReservedGb":44.451233792,"vramReservedPct":87.10133090882941,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.418120704,"vramAllocatedPct":98.79333014218277,"vramReservedGb":50.644123648,"vramReservedPct":99.23617852978492,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 156.25 MiB is free. Process 1415479 has 47.37 GiB memory in use. Of the allocated memory 46.96 GiB is allocated by PyTorch, and 95.53 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":28432.732026232403,"meanTps":28432.226086415878,"stepMs":576.2372741699219,"jitter":0.0010943926650698493,"achievedTflops":68.5959303738759,"nominalPeakTflops":154.8,"mfuNominalPct":44.31261652059166,"configuredPeakTflops":180.6,"mfuConfiguredPct":37.98224273193571,"vramAllocatedGb":22.085157376,"vramAllocatedPct":43.27543774784389,"vramReservedGb":23.261609984,"vramReservedPct":45.58067382717191,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":29351.584166605753,"meanTps":29342.341291649016,"stepMs":1116.3963012695312,"jitter":0.0012813347214813463,"achievedTflops":70.81272464418328,"nominalPeakTflops":154.8,"mfuNominalPct":45.744654162909086,"configuredPeakTflops":180.6,"mfuConfiguredPct":39.20970356820779,"vramAllocatedGb":42.026139136,"vramAllocatedPct":82.34940494191721,"vramReservedGb":44.451233792,"vramReservedPct":87.10133090882941,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.418120704,"vramAllocatedPct":98.79333014218277,"vramReservedGb":50.644123648,"vramReservedPct":99.23617852978492,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 156.25 MiB is free. Process 1449495 has 47.37 GiB memory in use. Of the allocated memory 46.96 GiB is allocated by PyTorch, and 95.53 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":76076.75638124115,"meanTps":76049.373832133,"stepMs":107.68072128295898,"jitter":0.0006154596868943814,"achievedTflops":183.54043076068916,"nominalPeakTflops":468.0,"mfuNominalPct":39.218040760831016,"configuredPeakTflops":468.0,"mfuConfiguredPct":39.218040760831016,"vramAllocatedGb":12.11460096,"vramAllocatedPct":11.880093793440741,"vramReservedGb":12.180258816,"vramReservedPct":11.94448068411355,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":83844.01061264377,"meanTps":83809.70415113073,"stepMs":195.4104995727539,"jitter":0.00031778875013689417,"achievedTflops":202.2794682180083,"nominalPeakTflops":468.0,"mfuNominalPct":43.222108593591514,"configuredPeakTflops":468.0,"mfuConfiguredPct":43.222108593591514,"vramAllocatedGb":22.085157376,"vramAllocatedPct":21.657646168956408,"vramReservedGb":23.263707136,"vramReservedPct":22.81338227081123,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":78737.6063130772,"meanTps":78729.71369126783,"stepMs":416.1670837402344,"jitter":0.00018395292265587285,"achievedTflops":189.95991505404368,"nominalPeakTflops":468.0,"mfuNominalPct":40.58972543889822,"configuredPeakTflops":468.0,"mfuConfiguredPct":40.58972543889822,"vramAllocatedGb":42.026139136,"vramAllocatedPct":41.21262238520076,"vramReservedGb":44.367347712,"vramReservedPct":43.50851125225659,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":75955.08447206994,"meanTps":75868.04856449865,"stepMs":107.85321426391602,"jitter":0.0008854779811165203,"achievedTflops":183.24688887374475,"nominalPeakTflops":468.0,"mfuNominalPct":39.15531813541555,"configuredPeakTflops":468.0,"mfuConfiguredPct":39.15531813541555,"vramAllocatedGb":12.133338112,"vramAllocatedPct":11.898468242910182,"vramReservedGb":13.176406016,"vramReservedPct":12.92134506513179,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":83637.83245853402,"meanTps":83608.49327042865,"stepMs":195.8922119140625,"jitter":0.0002713328299316591,"achievedTflops":201.78204917678218,"nominalPeakTflops":468.0,"mfuNominalPct":43.1158224736714,"configuredPeakTflops":468.0,"mfuConfiguredPct":43.1158224736714,"vramAllocatedGb":22.1011328,"vramAllocatedPct":21.67331235029714,"vramReservedGb":23.330816,"vramReservedPct":22.879192081742982,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":78652.28072840734,"meanTps":78642.2532593974,"stepMs":416.6185607910156,"jitter":0.00032080855233505834,"achievedTflops":189.7540611860537,"nominalPeakTflops":468.0,"mfuNominalPct":40.54573956966959,"configuredPeakTflops":468.0,"mfuConfiguredPct":40.54573956966959,"vramAllocatedGb":42.047604224,"vramAllocatedPct":41.23367196492414,"vramReservedGb":48.641343488,"vramReservedPct":47.69977358597274,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":39998.44974327412,"meanTps":39986.17203403294,"stepMs":409.6158752441406,"jitter":0.001087293586870738,"achievedTflops":156.89466020926957,"nominalPeakTflops":312.0,"mfuNominalPct":50.28675006707358,"configuredPeakTflops":312.0,"mfuConfiguredPct":50.28675006707358,"vramAllocatedGb":22.106317824,"vramAllocatedPct":26.015316983828523,"vramReservedGb":22.215131136,"vramReservedPct":26.14337145342757,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":40603.2762168039,"meanTps":40579.7237591037,"stepMs":807.0284729003906,"jitter":0.0021941824430739623,"achievedTflops":159.26710325791504,"nominalPeakTflops":312.0,"mfuNominalPct":51.04714848010097,"configuredPeakTflops":312.0,"mfuConfiguredPct":51.04714848010097,"vramAllocatedGb":42.047750656,"vramAllocatedPct":49.48293833834386,"vramReservedGb":44.430262272,"vramReservedPct":52.28674290685514,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":34398.88044126755,"meanTps":34372.02848550052,"stepMs":1905.1782836914062,"jitter":0.023762064994212132,"achievedTflops":134.9302458733288,"nominalPeakTflops":312.0,"mfuNominalPct":43.24687367734897,"configuredPeakTflops":312.0,"mfuConfiguredPct":43.24687367734897,"vramAllocatedGb":81.930354176,"vramAllocatedPct":96.41787254917462,"vramReservedGb":82.745229312,"vramReservedPct":97.37683980613973,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":39938.16888340845,"meanTps":39890.800577108275,"stepMs":410.234130859375,"jitter":0.003909890867448383,"achievedTflops":156.65820741956273,"nominalPeakTflops":312.0,"mfuNominalPct":50.21096391652652,"configuredPeakTflops":312.0,"mfuConfiguredPct":50.21096391652652,"vramAllocatedGb":22.106317824,"vramAllocatedPct":26.015316983828523,"vramReservedGb":22.215131136,"vramReservedPct":26.14337145342757,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":40591.972659382875,"meanTps":40612.38058531194,"stepMs":807.2532043457031,"jitter":0.002117569787441612,"achievedTflops":159.22276484449878,"nominalPeakTflops":312.0,"mfuNominalPct":51.032937450159864,"configuredPeakTflops":312.0,"mfuConfiguredPct":51.032937450159864,"vramAllocatedGb":42.047750656,"vramAllocatedPct":49.48293833834386,"vramReservedGb":44.430262272,"vramReservedPct":52.28674290685514,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":34028.03633648345,"meanTps":34038.169703512685,"stepMs":1925.9412841796875,"jitter":0.019514313161475994,"achievedTflops":133.47560300131357,"nominalPeakTflops":312.0,"mfuNominalPct":42.780641987600504,"configuredPeakTflops":312.0,"mfuConfiguredPct":42.780641987600504,"vramAllocatedGb":81.930354176,"vramAllocatedPct":96.41787254917462,"vramReservedGb":82.745229312,"vramReservedPct":97.37683980613973,"warmupSteps":13,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":101543.67170054436,"meanTps":101503.34705758841,"stepMs":645.3971862792969,"jitter":0.0003756909644358018,"achievedTflops":398.30693364653456,"nominalPeakTflops":2250.0,"mfuNominalPct":17.702530384290423,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.702530384290423,"vramAllocatedGb":81.930354176,"vramAllocatedPct":42.782802808780936,"vramReservedGb":86.566240256,"vramReservedPct":45.20359302748978,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":8,"tokensPerStep":131072,"status":"complete","stable":true,"tps":104154.98612751723,"meanTps":104071.08783528094,"stepMs":1258.4323120117188,"jitter":0.0007262657594726588,"achievedTflops":408.5498628687693,"nominalPeakTflops":2250.0,"mfuNominalPct":18.157771683056414,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.157771683056414,"vramAllocatedGb":161.695561216,"vramAllocatedPct":84.4349982388546,"vramReservedGb":171.234557952,"vramReservedPct":89.41611934414381,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.338528256,"vramAllocatedPct":99.39192626699803,"vramReservedGb":190.486413312,"vramReservedPct":99.46914962643356,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 192.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 177.81 MiB is free. Including non-PyTorch memory, this process has 178.16 GiB memory in use. Of the allocated memory 177.27 GiB is allocated by PyTorch, and 81.03 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":101614.74502454222,"meanTps":101630.4789680523,"stepMs":644.9457702636719,"jitter":0.0007694462789390705,"achievedTflops":398.5857200767627,"nominalPeakTflops":2250.0,"mfuNominalPct":17.71492089230056,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.71492089230056,"vramAllocatedGb":81.930354176,"vramAllocatedPct":42.782802808780936,"vramReservedGb":86.566240256,"vramReservedPct":45.20359302748978,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":8,"tokensPerStep":131072,"status":"complete","stable":true,"tps":104234.0989003959,"meanTps":104230.60861155041,"stepMs":1257.4771728515625,"jitter":0.0001738604808560667,"achievedTflops":408.8601841861873,"nominalPeakTflops":2250.0,"mfuNominalPct":18.171563741608324,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.171563741608324,"vramAllocatedGb":161.695561216,"vramAllocatedPct":84.4349982388546,"vramReservedGb":171.234557952,"vramReservedPct":89.41611934414381,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.338528256,"vramAllocatedPct":99.39192626699803,"vramReservedGb":190.486413312,"vramReservedPct":99.46914962643356,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 192.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 177.81 MiB is free. Including non-PyTorch memory, this process has 178.16 GiB memory in use. Of the allocated memory 177.27 GiB is allocated by PyTorch, and 81.03 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":102807.84352024396,"meanTps":102789.77734645682,"stepMs":637.4610900878906,"jitter":0.00044261536528180214,"achievedTflops":403.2656710318817,"nominalPeakTflops":2250.0,"mfuNominalPct":17.922918712528077,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.922918712528077,"vramAllocatedGb":81.930354176,"vramAllocatedPct":28.50458950683716,"vramReservedGb":86.566240256,"vramReservedPct":30.117471948758418,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":8,"tokensPerStep":131072,"status":"complete","stable":true,"tps":105861.62097337039,"meanTps":105860.79749758253,"stepMs":1238.1446533203125,"jitter":0.00014522137250747594,"achievedTflops":415.24416967215876,"nominalPeakTflops":2250.0,"mfuNominalPct":18.455296429873723,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.455296429873723,"vramAllocatedGb":161.695561216,"vramAllocatedPct":56.25589738863695,"vramReservedGb":171.234557952,"vramReservedPct":59.574633026989524,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":270.89739776,"vramAllocatedPct":94.24857489364001,"vramReservedGb":272.432627712,"vramReservedPct":94.7827004936143,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 31.25 GiB. GPU 0 has a total capacity of 267.69 GiB of which 14.40 GiB is free. Including non-PyTorch memory, this process has 253.27 GiB memory in use. Of the allocated memory 252.29 GiB is allocated by PyTorch, and 164.11 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":102811.57929076084,"meanTps":102810.08347337515,"stepMs":637.4379272460938,"jitter":0.00018539983069353316,"achievedTflops":403.2803246609505,"nominalPeakTflops":2250.0,"mfuNominalPct":17.923569984931135,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.923569984931135,"vramAllocatedGb":81.930354176,"vramAllocatedPct":28.50458950683716,"vramReservedGb":86.566240256,"vramReservedPct":30.117471948758418,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":8,"tokensPerStep":131072,"status":"complete","stable":true,"tps":105834.65332369422,"meanTps":105833.46672887109,"stepMs":1238.4601440429688,"jitter":9.871368642473822e-05,"achievedTflops":415.13838856664734,"nominalPeakTflops":2250.0,"mfuNominalPct":18.450595047406548,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.450595047406548,"vramAllocatedGb":161.695561216,"vramAllocatedPct":56.25589738863695,"vramReservedGb":171.234557952,"vramReservedPct":59.574633026989524,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":270.89739776,"vramAllocatedPct":94.24857489364001,"vramReservedGb":272.432627712,"vramReservedPct":94.7827004936143,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 31.25 GiB. GPU 0 has a total capacity of 267.69 GiB of which 14.40 GiB is free. Including non-PyTorch memory, this process has 253.27 GiB memory in use. Of the allocated memory 252.29 GiB is allocated by PyTorch, and 164.11 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":35731.44748388961,"meanTps":35733.986768909344,"stepMs":458.5316619873047,"jitter":0.0006762160323625054,"achievedTflops":140.157264787816,"nominalPeakTflops":165.2,"mfuNominalPct":84.84095931465859,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":68.19751753522796,"vramAllocatedGb":22.106317824,"vramAllocatedPct":87.5114840525304,"vramReservedGb":22.215131136,"vramReservedPct":87.94223939105416,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.629965824,"vramAllocatedPct":97.501758482876,"vramReservedGb":24.754782208,"vramReservedPct":97.99586460606092,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 23.69 MiB is free. Including non-PyTorch memory, this process has 23.49 GiB memory in use. Of the allocated memory 22.94 GiB is allocated by PyTorch, and 99.03 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":35744.02733977025,"meanTps":35747.35398900811,"stepMs":458.3702850341797,"jitter":0.0012773463232306655,"achievedTflops":140.20660950558735,"nominalPeakTflops":165.2,"mfuNominalPct":84.87082899853957,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":68.22152761605093,"vramAllocatedGb":22.106317824,"vramAllocatedPct":87.5114840525304,"vramReservedGb":22.215131136,"vramReservedPct":87.94223939105416,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.629965824,"vramAllocatedPct":97.501758482876,"vramReservedGb":24.754782208,"vramReservedPct":97.99586460606092,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 23.69 MiB is free. Including non-PyTorch memory, this process has 23.49 GiB memory in use. Of the allocated memory 22.94 GiB is allocated by PyTorch, and 99.03 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":53491.08115676321,"meanTps":53494.60503983449,"stepMs":306.29405212402344,"jitter":0.0019009181804654112,"achievedTflops":209.81975692015538,"nominalPeakTflops":209.5,"mfuNominalPct":100.15262860150615,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":78.01533237664249,"vramAllocatedGb":22.106317824,"vramAllocatedPct":65.65805682671981,"vramReservedGb":22.215131136,"vramReservedPct":65.9812436495856,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_extension_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":31.663030272,"vramAllocatedPct":94.04248402907274,"vramReservedGb":31.843155968,"vramReservedPct":94.57747602901044,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 1.22 GiB is free. Including non-PyTorch memory, this process has 30.12 GiB memory in use. Of the allocated memory 29.39 GiB is allocated by PyTorch, and 147.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":53273.41249727615,"meanTps":53288.200774456775,"stepMs":307.5455322265625,"jitter":0.0029500021632023017,"achievedTflops":208.96594757035243,"nominalPeakTflops":209.5,"mfuNominalPct":99.74508237248327,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":77.69786837235185,"vramAllocatedGb":22.106317824,"vramAllocatedPct":65.65805682671981,"vramReservedGb":22.215131136,"vramReservedPct":65.9812436495856,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_extension_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":31.663030272,"vramAllocatedPct":94.04248402907274,"vramReservedGb":31.843155968,"vramReservedPct":94.57747602901044,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 1.22 GiB is free. Including non-PyTorch memory, this process has 30.12 GiB memory in use. Of the allocated memory 29.39 GiB is allocated by PyTorch, and 147.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":76835.2044234516,"meanTps":76779.80191375337,"stepMs":213.23558807373047,"jitter":0.0006346448871640938,"achievedTflops":301.3875129536565,"nominalPeakTflops":989.5,"mfuNominalPct":30.458566240895045,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.458566240895045,"vramAllocatedGb":22.156387328,"vramAllocatedPct":26.060974529856683,"vramReservedGb":22.257074176,"vramReservedPct":26.179405271401965,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":80463.23978456594,"meanTps":80455.77300492807,"stepMs":407.2418670654297,"jitter":0.0007547710268532698,"achievedTflops":315.6185488778667,"nominalPeakTflops":989.5,"mfuNominalPct":31.896770983109317,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.896770983109317,"vramAllocatedGb":42.09782016,"vramAllocatedPct":49.5166564255618,"vramReservedGb":44.430262272,"vramReservedPct":52.26014134362782,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":75805.82462856981,"meanTps":75122.30723613482,"stepMs":864.5245971679688,"jitter":0.002349970401085936,"achievedTflops":297.3497516358838,"nominalPeakTflops":989.5,"mfuNominalPct":30.05050547103424,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.05050547103424,"vramAllocatedGb":81.98042368,"vramAllocatedPct":96.4277118757246,"vramReservedGb":82.241912832,"vramReservedPct":96.73528287226038,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":76796.2062421004,"meanTps":76652.92495618292,"stepMs":213.3438720703125,"jitter":0.0005308821371129012,"achievedTflops":301.23454186474805,"nominalPeakTflops":989.5,"mfuNominalPct":30.44310680795837,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.44310680795837,"vramAllocatedGb":22.156387328,"vramAllocatedPct":26.060974529856683,"vramReservedGb":22.257074176,"vramReservedPct":26.179405271401965,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":80394.34326493366,"meanTps":80351.52381211873,"stepMs":407.5908660888672,"jitter":0.0005362738931654928,"achievedTflops":315.3483010030943,"nominalPeakTflops":989.5,"mfuNominalPct":31.869459424264207,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.869459424264207,"vramAllocatedGb":42.09782016,"vramAllocatedPct":49.5166564255618,"vramReservedGb":44.430262272,"vramReservedPct":52.26014134362782,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":75727.84794592776,"meanTps":75326.95657524504,"stepMs":865.414794921875,"jitter":0.003423196940592805,"achievedTflops":297.04388665346835,"nominalPeakTflops":989.5,"mfuNominalPct":30.019594406616307,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.019594406616307,"vramAllocatedGb":81.98042368,"vramAllocatedPct":96.4277118757246,"vramReservedGb":82.241912832,"vramReservedPct":96.73528287226038,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":85706.58837448226,"meanTps":85697.06451219587,"stepMs":382.3276672363281,"jitter":0.0008360851310967581,"achievedTflops":336.18568087057605,"nominalPeakTflops":989.5,"mfuNominalPct":33.97530882977019,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.97530882977019,"vramAllocatedGb":42.09782016,"vramAllocatedPct":28.04263737901418,"vramReservedGb":44.488982528,"vramReservedPct":29.63546330076777,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":88462.40613645007,"meanTps":88423.7484403229,"stepMs":740.83447265625,"jitter":0.001304224798959155,"achievedTflops":346.99542710168623,"nominalPeakTflops":989.5,"mfuNominalPct":35.06775412851806,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.06775412851806,"vramAllocatedGb":81.98042368,"vramAllocatedPct":54.60965163276015,"vramReservedGb":86.68577792,"vramReservedPct":57.744031089716024,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.581341184,"vramAllocatedPct":90.98098212699304,"vramReservedGb":137.545908224,"vramReservedPct":91.62350954593455,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.74 GiB is free. Including non-PyTorch memory, this process has 128.06 GiB memory in use. Of the allocated memory 127.20 GiB is allocated by PyTorch, and 139.88 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":85563.85688227315,"meanTps":85545.73748137648,"stepMs":382.96543884277344,"jitter":0.0009281452646688546,"achievedTflops":335.6258139478567,"nominalPeakTflops":989.5,"mfuNominalPct":33.91872803919724,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.91872803919724,"vramAllocatedGb":42.09782016,"vramAllocatedPct":28.04263737901418,"vramReservedGb":44.488982528,"vramReservedPct":29.63546330076777,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":88306.14037259184,"meanTps":88321.33635299874,"stepMs":742.1454467773438,"jitter":0.0007538399585305449,"achievedTflops":346.3824717476603,"nominalPeakTflops":989.5,"mfuNominalPct":35.00580816045076,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.00580816045076,"vramAllocatedGb":81.98042368,"vramAllocatedPct":54.60965163276015,"vramReservedGb":86.68577792,"vramReservedPct":57.744031089716024,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.581341184,"vramAllocatedPct":90.98098212699304,"vramReservedGb":137.545908224,"vramReservedPct":91.62350954593455,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.74 GiB is free. Including non-PyTorch memory, this process has 128.06 GiB memory in use. Of the allocated memory 127.20 GiB is allocated by PyTorch, and 139.88 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23105.364492132798,"meanTps":23106.132781668948,"stepMs":709.0993957519531,"jitter":0.001321885417263358,"achievedTflops":90.6312203165842,"nominalPeakTflops":154.8,"mfuNominalPct":58.547299946113824,"configuredPeakTflops":180.6,"mfuConfiguredPct":50.18339995381185,"vramAllocatedGb":22.106317824,"vramAllocatedPct":43.31690123549535,"vramReservedGb":22.215131136,"vramReservedPct":43.530118810965746,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":23561.535607209807,"meanTps":23562.497200183465,"stepMs":1390.7412719726562,"jitter":0.0007275740507124909,"achievedTflops":92.4205599674122,"nominalPeakTflops":154.8,"mfuNominalPct":59.70320411331538,"configuredPeakTflops":180.6,"mfuConfiguredPct":51.17417495427033,"vramAllocatedGb":42.047750656,"vramAllocatedPct":82.39175229640844,"vramReservedGb":44.430262272,"vramReservedPct":87.06023762193149,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.439092224,"vramAllocatedPct":98.83442342908069,"vramReservedGb":50.654609408,"vramReservedPct":99.25672517323389,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 126.25 MiB is free. Process 1472317 has 47.40 GiB memory in use. Of the allocated memory 46.98 GiB is allocated by PyTorch, and 105.53 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23074.66780821329,"meanTps":23067.332400713534,"stepMs":710.042724609375,"jitter":0.0011772644399630717,"achievedTflops":90.51081200516178,"nominalPeakTflops":154.8,"mfuNominalPct":58.469516799200115,"configuredPeakTflops":180.6,"mfuConfiguredPct":50.116728685028676,"vramAllocatedGb":22.106317824,"vramAllocatedPct":43.31690123549535,"vramReservedGb":22.215131136,"vramReservedPct":43.530118810965746,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":23559.609348313534,"meanTps":23559.277687209993,"stepMs":1390.85498046875,"jitter":0.0003422870813850659,"achievedTflops":92.41300418120184,"nominalPeakTflops":154.8,"mfuNominalPct":59.69832311447148,"configuredPeakTflops":180.6,"mfuConfiguredPct":51.169991240975556,"vramAllocatedGb":42.047750656,"vramAllocatedPct":82.39175229640844,"vramReservedGb":44.430262272,"vramReservedPct":87.06023762193149,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.439092224,"vramAllocatedPct":98.83442342908069,"vramReservedGb":50.654609408,"vramReservedPct":99.25672517323389,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 126.25 MiB is free. Process 1479887 has 47.40 GiB memory in use. Of the allocated memory 46.98 GiB is allocated by PyTorch, and 105.53 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":68742.68646671713,"meanTps":68649.58311260206,"stepMs":238.33808135986328,"jitter":0.0020924297160298828,"achievedTflops":269.64446133019277,"nominalPeakTflops":468.0,"mfuNominalPct":57.61633789106683,"configuredPeakTflops":468.0,"mfuConfiguredPct":57.61633789106683,"vramAllocatedGb":22.120400384,"vramAllocatedPct":21.692206963982624,"vramReservedGb":23.21547264,"vramReservedPct":22.76608146920403,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":65575.7040395552,"meanTps":65558.10045033645,"stepMs":499.697265625,"jitter":0.00028571646773437043,"achievedTflops":257.2219140817997,"nominalPeakTflops":468.0,"mfuNominalPct":54.961947453376006,"configuredPeakTflops":468.0,"mfuConfiguredPct":54.961947453376006,"vramAllocatedGb":42.072114688,"vramAllocatedPct":41.25770797008867,"vramReservedGb":44.472205312,"vramReservedPct":43.61133908183746,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":68813.95113425641,"meanTps":68816.90738315192,"stepMs":238.09125518798828,"jitter":0.0006575640983203276,"achievedTflops":269.9239983090074,"nominalPeakTflops":468.0,"mfuNominalPct":57.67606801474517,"configuredPeakTflops":468.0,"mfuConfiguredPct":57.67606801474517,"vramAllocatedGb":22.120400384,"vramAllocatedPct":21.692206963982624,"vramReservedGb":23.21547264,"vramReservedPct":22.76608146920403,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":65554.56125547267,"meanTps":65550.54142580196,"stepMs":499.8584289550781,"jitter":0.000492182662070883,"achievedTflops":257.1389811195025,"nominalPeakTflops":468.0,"mfuNominalPct":54.94422673493643,"configuredPeakTflops":468.0,"mfuConfiguredPct":54.94422673493643,"vramAllocatedGb":42.072114688,"vramAllocatedPct":41.25770797008867,"vramReservedGb":44.472205312,"vramReservedPct":43.61133908183746,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":27978.889776692493,"meanTps":27983.84546692836,"stepMs":1171.168701171875,"jitter":0.004187162137513296,"achievedTflops":194.2411314422232,"nominalPeakTflops":312.0,"mfuNominalPct":62.25677289814846,"configuredPeakTflops":312.0,"mfuConfiguredPct":62.25677289814846,"vramAllocatedGb":42.089751552,"vramAllocatedPct":49.53236613684672,"vramReservedGb":42.328915968,"vramReservedPct":49.813821336352504,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":24641.42747183799,"meanTps":24541.32040609314,"stepMs":2659.586181640625,"jitter":0.014394006287140278,"achievedTflops":171.07107503845748,"nominalPeakTflops":312.0,"mfuNominalPct":54.83047276873638,"configuredPeakTflops":312.0,"mfuConfiguredPct":54.83047276873638,"vramAllocatedGb":81.972937216,"vramAllocatedPct":96.46798543059407,"vramReservedGb":82.325798912,"vramReservedPct":96.88324268627892,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":28054.241836277568,"meanTps":28068.61049023206,"stepMs":1168.0230102539062,"jitter":0.005505957598437238,"achievedTflops":194.7642568924153,"nominalPeakTflops":312.0,"mfuNominalPct":62.424441311671565,"configuredPeakTflops":312.0,"mfuConfiguredPct":62.424441311671565,"vramAllocatedGb":42.089751552,"vramAllocatedPct":49.53236613684672,"vramReservedGb":42.328915968,"vramReservedPct":49.813821336352504,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":24386.721221871365,"meanTps":23363.445350129816,"stepMs":2687.3641357421875,"jitter":0.020795364800191795,"achievedTflops":169.3027979347628,"nominalPeakTflops":312.0,"mfuNominalPct":54.26371728678294,"configuredPeakTflops":312.0,"mfuConfiguredPct":54.26371728678294,"vramAllocatedGb":81.972937216,"vramAllocatedPct":96.46798543059407,"vramReservedGb":82.325798912,"vramReservedPct":96.88324268627892,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":65232.143388724326,"meanTps":65214.97744255213,"stepMs":1004.6580810546875,"jitter":0.00034492752205637824,"achievedTflops":452.8687678230319,"nominalPeakTflops":2250.0,"mfuNominalPct":20.12750079213475,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.12750079213475,"vramAllocatedGb":81.972937216,"vramAllocatedPct":42.80503903394609,"vramReservedGb":86.729818112,"vramReservedPct":45.289010931849106,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":4,"tokensPerStep":131072,"status":"complete","stable":true,"tps":66324.62710064974,"meanTps":66320.42297773864,"stepMs":1976.2191772460938,"jitter":0.00023046016978924976,"achievedTflops":460.4532457626593,"nominalPeakTflops":2250.0,"mfuNominalPct":20.464588700562636,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.464588700562636,"vramAllocatedGb":161.738784256,"vramAllocatedPct":84.45756866242611,"vramReservedGb":171.121311744,"vramReservedPct":89.35698387189504,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.380602368,"vramAllocatedPct":99.41389673759045,"vramReservedGb":190.467538944,"vramReservedPct":99.45929371439209,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 192.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 135.81 MiB is free. Including non-PyTorch memory, this process has 178.21 GiB memory in use. Of the allocated memory 177.31 GiB is allocated by PyTorch, and 82.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":64989.879145141276,"meanTps":64930.24311722012,"stepMs":1008.4031677246094,"jitter":0.0027737149995816102,"achievedTflops":451.186868321045,"nominalPeakTflops":2250.0,"mfuNominalPct":20.052749703157556,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.052749703157556,"vramAllocatedGb":81.972937216,"vramAllocatedPct":42.80503903394609,"vramReservedGb":86.729818112,"vramReservedPct":45.289010931849106,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":4,"tokensPerStep":131072,"status":"complete","stable":true,"tps":66242.2232040301,"meanTps":66219.15971788554,"stepMs":1978.6775512695312,"jitter":0.0007337848676192684,"achievedTflops":459.88116351628014,"nominalPeakTflops":2250.0,"mfuNominalPct":20.439162822945782,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.439162822945782,"vramAllocatedGb":161.738784256,"vramAllocatedPct":84.45756866242611,"vramReservedGb":171.121311744,"vramReservedPct":89.35698387189504,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.380602368,"vramAllocatedPct":99.41389673759045,"vramReservedGb":190.467538944,"vramReservedPct":99.45929371439209,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 192.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 135.75 MiB is free. Including non-PyTorch memory, this process has 178.21 GiB memory in use. Of the allocated memory 177.31 GiB is allocated by PyTorch, and 82.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":65927.01107465562,"meanTps":65912.73064072116,"stepMs":994.0690307617188,"jitter":0.00028674646080541037,"achievedTflops":457.6928293420979,"nominalPeakTflops":2250.0,"mfuNominalPct":20.341903526315463,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.341903526315463,"vramAllocatedGb":81.972937216,"vramAllocatedPct":28.519404676225367,"vramReservedGb":86.729818112,"vramReservedPct":30.174382719919887,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":4,"tokensPerStep":131072,"status":"complete","stable":true,"tps":67281.67909530878,"meanTps":67280.70016974666,"stepMs":1948.1083374023438,"jitter":6.944358677273787e-05,"achievedTflops":467.0975001907413,"nominalPeakTflops":2250.0,"mfuNominalPct":20.759888897366277,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.759888897366277,"vramAllocatedGb":161.738784256,"vramAllocatedPct":56.27093522198735,"vramReservedGb":171.121311744,"vramReservedPct":59.535233262339275,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":270.9403648,"vramAllocatedPct":94.26352366140554,"vramReservedGb":272.956915712,"vramReservedPct":94.96510681143953,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 31.25 GiB. GPU 0 has a total capacity of 267.69 GiB of which 14.36 GiB is free. Including non-PyTorch memory, this process has 253.31 GiB memory in use. Of the allocated memory 252.33 GiB is allocated by PyTorch, and 163.13 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":65946.34533779895,"meanTps":65942.8236329002,"stepMs":993.777587890625,"jitter":0.00016358447325926732,"achievedTflops":457.82705586711506,"nominalPeakTflops":2250.0,"mfuNominalPct":20.347869149649558,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.347869149649558,"vramAllocatedGb":81.972937216,"vramAllocatedPct":28.519404676225367,"vramReservedGb":86.729818112,"vramReservedPct":30.174382719919887,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":4,"tokensPerStep":131072,"status":"complete","stable":true,"tps":67289.38881342602,"meanTps":67289.90306830539,"stepMs":1947.8851318359375,"jitter":0.00013774169138140916,"achievedTflops":467.15102427200924,"nominalPeakTflops":2250.0,"mfuNominalPct":20.762267745422633,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.762267745422633,"vramAllocatedGb":161.738784256,"vramAllocatedPct":56.27093522198735,"vramReservedGb":171.121311744,"vramReservedPct":59.535233262339275,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":270.9403648,"vramAllocatedPct":94.26352366140554,"vramReservedGb":272.956915712,"vramReservedPct":94.96510681143953,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 31.25 GiB. GPU 0 has a total capacity of 267.69 GiB of which 14.36 GiB is free. Including non-PyTorch memory, this process has 253.31 GiB memory in use. Of the allocated memory 252.33 GiB is allocated by PyTorch, and 163.13 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.6591232,"vramAllocatedPct":97.6171827369355,"vramReservedGb":24.73590784,"vramReservedPct":97.92114732535484,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 96.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 29.69 MiB is free. Including non-PyTorch memory, this process has 23.49 GiB memory in use. Of the allocated memory 22.97 GiB is allocated by PyTorch, and 73.23 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.6591232,"vramAllocatedPct":97.6171827369355,"vramReservedGb":24.73590784,"vramReservedPct":97.92114732535484,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 96.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 29.69 MiB is free. Including non-PyTorch memory, this process has 23.49 GiB memory in use. Of the allocated memory 22.97 GiB is allocated by PyTorch, and 73.23 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.497822208,"vramAllocatedPct":96.52190266444897,"vramReservedGb":32.686211072,"vramReservedPct":97.08143713041075,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 31.36 GiB of which 427.88 MiB is free. Including non-PyTorch memory, this process has 30.93 GiB memory in use. Of the allocated memory 30.27 GiB is allocated by PyTorch, and 79.66 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.497822208,"vramAllocatedPct":96.52190266444897,"vramReservedGb":32.686211072,"vramReservedPct":97.08143713041075,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 31.36 GiB of which 427.88 MiB is free. Including non-PyTorch memory, this process has 30.93 GiB memory in use. Of the allocated memory 30.27 GiB is allocated by PyTorch, and 79.66 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_32k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":54547.12757846045,"meanTps":54552.61064617481,"stepMs":600.7282409667969,"jitter":0.00025811314118001405,"achievedTflops":378.68892805710067,"nominalPeakTflops":989.5,"mfuNominalPct":38.27073552876207,"configuredPeakTflops":989.5,"mfuConfiguredPct":38.27073552876207,"vramAllocatedGb":42.139821056,"vramAllocatedPct":49.56605907702673,"vramReservedGb":42.391830528,"vramReservedPct":49.86247980364829,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_32k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":52487.502453305315,"meanTps":52380.24698941658,"stepMs":1248.6019897460938,"jitter":0.003702895342667177,"achievedTflops":364.3901507342696,"nominalPeakTflops":989.5,"mfuNominalPct":36.82568476344311,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.82568476344311,"vramAllocatedGb":82.02300672,"vramAllocatedPct":96.47779926156126,"vramReservedGb":82.346770432,"vramReservedPct":96.85861937123053,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_32k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":54555.558406495074,"meanTps":54553.475123217606,"stepMs":600.6354064941406,"jitter":0.0004516736502547224,"achievedTflops":378.7474583844122,"nominalPeakTflops":989.5,"mfuNominalPct":38.27665067048127,"configuredPeakTflops":989.5,"mfuConfiguredPct":38.27665067048127,"vramAllocatedGb":42.139821056,"vramAllocatedPct":49.56605907702673,"vramReservedGb":42.391830528,"vramReservedPct":49.86247980364829,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_32k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":52087.76755003665,"meanTps":51006.31312616694,"stepMs":1258.18408203125,"jitter":0.007096297163376813,"achievedTflops":361.615024183232,"nominalPeakTflops":989.5,"mfuNominalPct":36.545227305025975,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.545227305025975,"vramAllocatedGb":82.02300672,"vramAllocatedPct":96.47779926156126,"vramReservedGb":82.346770432,"vramReservedPct":96.85861937123053,"warmupSteps":14,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":56944.50781088371,"meanTps":56947.70342920706,"stepMs":575.4374084472656,"jitter":0.0002947755170746523,"achievedTflops":395.3325423162711,"nominalPeakTflops":989.5,"mfuNominalPct":39.95275819264993,"configuredPeakTflops":989.5,"mfuConfiguredPct":39.95275819264993,"vramAllocatedGb":42.139821056,"vramAllocatedPct":28.07061545226465,"vramReservedGb":42.391830528,"vramReservedPct":28.238486620237566,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":58206.55045108455,"meanTps":58202.122922811395,"stepMs":1125.92138671875,"jitter":0.0015464063611963638,"achievedTflops":404.09416911123964,"nominalPeakTflops":989.5,"mfuNominalPct":40.83821820224756,"configuredPeakTflops":989.5,"mfuConfiguredPct":40.83821820224756,"vramAllocatedGb":82.02300672,"vramAllocatedPct":54.638017489820626,"vramReservedGb":86.813704192,"vramReservedPct":57.82924666722837,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.623796224,"vramAllocatedPct":91.00926271936353,"vramReservedGb":137.355067392,"vramReservedPct":91.49638466800631,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.72 GiB is free. Including non-PyTorch memory, this process has 128.08 GiB memory in use. Of the allocated memory 127.24 GiB is allocated by PyTorch, and 117.39 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":56892.02672109692,"meanTps":56888.40784550284,"stepMs":575.9682312011719,"jitter":0.0004908135592627627,"achievedTflops":394.96819668494436,"nominalPeakTflops":989.5,"mfuNominalPct":39.915937007068656,"configuredPeakTflops":989.5,"mfuConfiguredPct":39.915937007068656,"vramAllocatedGb":42.139821056,"vramAllocatedPct":28.07061545226465,"vramReservedGb":42.391830528,"vramReservedPct":28.238486620237566,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":58042.2257637334,"meanTps":58043.38165376013,"stepMs":1129.1090087890625,"jitter":0.00047693383573733125,"achievedTflops":402.9533585412088,"nominalPeakTflops":989.5,"mfuNominalPct":40.72292658324495,"configuredPeakTflops":989.5,"mfuConfiguredPct":40.72292658324495,"vramAllocatedGb":82.02300672,"vramAllocatedPct":54.638017489820626,"vramReservedGb":86.813704192,"vramReservedPct":57.82924666722837,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":136.623796224,"vramAllocatedPct":91.00926271936353,"vramReservedGb":137.355067392,"vramReservedPct":91.49638466800631,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 139.81 GiB of which 11.72 GiB is free. Including non-PyTorch memory, this process has 128.08 GiB memory in use. Of the allocated memory 127.24 GiB is allocated by PyTorch, and 117.39 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":16787.530363577447,"meanTps":16786.57582747505,"stepMs":1951.9249877929688,"jitter":0.0006074052310702522,"achievedTflops":116.54604303342865,"nominalPeakTflops":154.8,"mfuNominalPct":75.28814149446295,"configuredPeakTflops":180.6,"mfuConfiguredPct":64.53269270953967,"vramAllocatedGb":42.089751552,"vramAllocatedPct":82.47405223791472,"vramReservedGb":42.328915968,"vramReservedPct":82.94269027475998,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.481035264,"vramAllocatedPct":98.91661000287652,"vramReservedGb":50.577014784,"vramReservedPct":99.1046800117116,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 100.25 MiB is free. Process 1490581 has 47.42 GiB memory in use. Of the allocated memory 47.01 GiB is allocated by PyTorch, and 91.53 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":16799.125209108654,"meanTps":16803.141619714876,"stepMs":1950.5777587890625,"jitter":0.0007745103254350532,"achievedTflops":116.62653929088756,"nominalPeakTflops":154.8,"mfuNominalPct":75.34014166078008,"configuredPeakTflops":180.6,"mfuConfiguredPct":64.57726428066864,"vramAllocatedGb":42.089751552,"vramAllocatedPct":82.47405223791472,"vramReservedGb":42.328915968,"vramReservedPct":82.94269027475998,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.481035264,"vramAllocatedPct":98.91661000287652,"vramReservedGb":50.577014784,"vramReservedPct":99.1046800117116,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 100.25 MiB is free. Process 1493539 has 47.42 GiB memory in use. Of the allocated memory 47.01 GiB is allocated by PyTorch, and 91.53 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":49531.700968236226,"meanTps":49530.9394237972,"stepMs":661.5561218261719,"jitter":0.00040912288849767576,"achievedTflops":343.86974304973364,"nominalPeakTflops":468.0,"mfuNominalPct":73.47644082259266,"configuredPeakTflops":468.0,"mfuConfiguredPct":73.47644082259266,"vramAllocatedGb":42.11674624,"vramAllocatedPct":41.30147557132106,"vramReservedGb":45.42431232,"vramReservedPct":44.54501577443173,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":49629.03654689281,"meanTps":49626.486335560105,"stepMs":1320.5172729492188,"jitter":0.00018241813919854793,"achievedTflops":344.5454872654168,"nominalPeakTflops":468.0,"mfuNominalPct":73.62083061226855,"configuredPeakTflops":468.0,"mfuConfiguredPct":73.62083061226855,"vramAllocatedGb":82.012898816,"vramAllocatedPct":80.42534239658895,"vramReservedGb":95.082774528,"vramReservedPct":93.24221930733889,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":49516.64807347529,"meanTps":49515.63764164318,"stepMs":661.7572326660156,"jitter":0.0004223310046765746,"achievedTflops":343.7652395711045,"nominalPeakTflops":468.0,"mfuNominalPct":73.45411101946677,"configuredPeakTflops":468.0,"mfuConfiguredPct":73.45411101946677,"vramAllocatedGb":42.11674624,"vramAllocatedPct":41.30147557132106,"vramReservedGb":45.42431232,"vramReservedPct":44.54501577443173,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"150m","modelLabel":"150M","parameters":150436608,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":49623.33001491877,"meanTps":49621.615330059365,"stepMs":1320.6691284179688,"jitter":0.0002803791117256928,"achievedTflops":344.50587013850094,"nominalPeakTflops":468.0,"mfuNominalPct":73.61236541420959,"configuredPeakTflops":468.0,"mfuConfiguredPct":73.61236541420959,"vramAllocatedGb":82.012898816,"vramAllocatedPct":80.42534239658895,"vramReservedGb":95.082774528,"vramReservedPct":93.24221930733889,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":47264.46694365139,"meanTps":47237.49688835055,"stepMs":346.6451873779297,"jitter":0.00209849241599263,"achievedTflops":76.64001829484354,"nominalPeakTflops":312.0,"mfuNominalPct":24.56410842783447,"configuredPeakTflops":312.0,"mfuConfiguredPct":24.56410842783447,"vramAllocatedGb":30.534776832,"vramAllocatedPct":35.93415713278688,"vramReservedGb":31.702646784,"vramReservedPct":37.30853830467899,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":49986.60064924651,"meanTps":49986.345837618675,"stepMs":655.5356750488281,"jitter":0.0013753703261357821,"achievedTflops":81.05399755851622,"nominalPeakTflops":312.0,"mfuNominalPct":25.9788453713193,"configuredPeakTflops":312.0,"mfuConfiguredPct":25.9788453713193,"vramAllocatedGb":57.601193472,"vramAllocatedPct":67.78665351468143,"vramReservedGb":60.007907328,"vramReservedPct":70.61893993848545,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.225555968,"vramAllocatedPct":99.11892853851292,"vramReservedGb":84.376813568,"vramReservedPct":99.29693260239827,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 84.75 MiB is free. Process 764857 has 79.05 GiB memory in use. Of the allocated memory 78.42 GiB is allocated by PyTorch, and 128.25 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":47187.22340880285,"meanTps":47176.744768137156,"stepMs":347.21263122558594,"jitter":0.001623665591192624,"achievedTflops":76.51476678336437,"nominalPeakTflops":312.0,"mfuNominalPct":24.523963712616784,"configuredPeakTflops":312.0,"mfuConfiguredPct":24.523963712616784,"vramAllocatedGb":30.534776832,"vramAllocatedPct":35.93415713278688,"vramReservedGb":31.702646784,"vramReservedPct":37.30853830467899,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":50001.79863485552,"meanTps":50017.10791431464,"stepMs":655.33642578125,"jitter":0.001637593582658597,"achievedTflops":81.07864131249123,"nominalPeakTflops":312.0,"mfuNominalPct":25.986744010413855,"configuredPeakTflops":312.0,"mfuConfiguredPct":25.986744010413855,"vramAllocatedGb":57.601193472,"vramAllocatedPct":67.78665351468143,"vramReservedGb":60.007907328,"vramReservedPct":70.61893993848545,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.225555968,"vramAllocatedPct":99.11892853851292,"vramReservedGb":84.376813568,"vramReservedPct":99.29693260239827,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 84.75 MiB is free. Process 852916 has 79.05 GiB memory in use. Of the allocated memory 78.42 GiB is allocated by PyTorch, and 128.25 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":146013.0649201807,"meanTps":145716.8690786083,"stepMs":224.41827392578125,"jitter":0.0033907064407244316,"achievedTflops":236.76230137346212,"nominalPeakTflops":2250.0,"mfuNominalPct":10.522768949931649,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.522768949931649,"vramAllocatedGb":57.601193472,"vramAllocatedPct":30.078479785028186,"vramReservedGb":60.02049024,"vramReservedPct":31.341800291844507,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":158445.85734710182,"meanTps":158298.31968979398,"stepMs":413.6176300048828,"jitter":0.0006077125021373752,"achievedTflops":256.9222545194737,"nominalPeakTflops":2250.0,"mfuNominalPct":11.418766867532165,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.418766867532165,"vramAllocatedGb":111.734027264,"vramAllocatedPct":58.34583413612246,"vramReservedGb":116.417101824,"vramReservedPct":60.79126547172859,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.510797312,"vramAllocatedPct":97.39314254600453,"vramReservedGb":187.269382144,"vramReservedPct":97.7892641740335,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.80 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.72 GiB is free. Including non-PyTorch memory, this process has 174.62 GiB memory in use. Of the allocated memory 173.70 GiB is allocated by PyTorch, and 103.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":146111.82902819518,"meanTps":146042.66473590105,"stepMs":224.2665786743164,"jitter":0.002428896091381857,"achievedTflops":236.92244880629207,"nominalPeakTflops":2250.0,"mfuNominalPct":10.52988661361298,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.52988661361298,"vramAllocatedGb":57.601193472,"vramAllocatedPct":30.078479785028186,"vramReservedGb":60.02049024,"vramReservedPct":31.341800291844507,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":158251.9163458157,"meanTps":158162.12196838373,"stepMs":414.12452697753906,"jitter":0.000556110237761037,"achievedTflops":256.6077763745195,"nominalPeakTflops":2250.0,"mfuNominalPct":11.404790061089757,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.404790061089757,"vramAllocatedGb":111.734027264,"vramAllocatedPct":58.34583413612246,"vramReservedGb":116.417101824,"vramReservedPct":60.79126547172859,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.510797312,"vramAllocatedPct":97.39314254600453,"vramReservedGb":187.269382144,"vramReservedPct":97.7892641740335,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.80 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.72 GiB is free. Including non-PyTorch memory, this process has 174.62 GiB memory in use. Of the allocated memory 173.70 GiB is allocated by PyTorch, and 103.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":159413.6892825008,"meanTps":159414.2016843072,"stepMs":411.1064758300781,"jitter":0.00035719169807270167,"achievedTflops":258.4916080324149,"nominalPeakTflops":2250.0,"mfuNominalPct":11.488515912551772,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.488515912551772,"vramAllocatedGb":111.734027264,"vramAllocatedPct":38.87365815927401,"vramReservedGb":116.38145024,"vramReservedPct":40.49055443084327,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":256,"tokensPerStep":131072,"status":"complete","stable":true,"tps":167668.840626528,"meanTps":167648.33598684223,"stepMs":781.7314147949219,"jitter":0.00010748018359780252,"achievedTflops":271.87745560342887,"nominalPeakTflops":2250.0,"mfuNominalPct":12.083442471263504,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.083442471263504,"vramAllocatedGb":219.999694848,"vramAllocatedPct":76.54063083628965,"vramReservedGb":229.170479104,"vramReservedPct":79.7312607719481,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":512,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.044873216,"vramAllocatedPct":99.51857005331736,"vramReservedGb":286.179459072,"vramReservedPct":99.56539414699168,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 337.62 MiB is free. Including non-PyTorch memory, this process has 267.34 GiB memory in use. Of the allocated memory 266.40 GiB is allocated by PyTorch, and 128.35 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":159382.03444644937,"meanTps":159207.89941714713,"stepMs":411.18812561035156,"jitter":0.00030270813105051,"achievedTflops":258.44027925688897,"nominalPeakTflops":2250.0,"mfuNominalPct":11.48623463363951,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.48623463363951,"vramAllocatedGb":111.734027264,"vramAllocatedPct":38.87365815927401,"vramReservedGb":116.38145024,"vramReservedPct":40.49055443084327,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":256,"tokensPerStep":131072,"status":"complete","stable":true,"tps":167668.35625865846,"meanTps":167651.55173563692,"stepMs":781.7336730957031,"jitter":0.00013624535341381813,"achievedTflops":271.87667019390705,"nominalPeakTflops":2250.0,"mfuNominalPct":12.083407564173646,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.083407564173646,"vramAllocatedGb":219.999694848,"vramAllocatedPct":76.54063083628965,"vramReservedGb":229.170479104,"vramReservedPct":79.7312607719481,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":512,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.044873216,"vramAllocatedPct":99.51857005331736,"vramReservedGb":286.179459072,"vramReservedPct":99.56539414699168,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 337.62 MiB is free. Including non-PyTorch memory, this process has 267.34 GiB memory in use. Of the allocated memory 266.40 GiB is allocated by PyTorch, and 128.35 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":39965.7605371856,"meanTps":39956.023962841726,"stepMs":102.48772811889648,"jitter":0.0012783345069739712,"achievedTflops":64.80506005471119,"nominalPeakTflops":165.2,"mfuNominalPct":39.22824458517627,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":31.532751628275662,"vramAllocatedGb":10.234963968,"vramAllocatedPct":40.51678317460235,"vramReservedGb":10.554966016,"vramReservedPct":41.783563754854676,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":40872.86233302286,"meanTps":40870.94964948427,"stepMs":200.42638397216797,"jitter":0.0006616037274354863,"achievedTflops":66.27593876600893,"nominalPeakTflops":165.2,"mfuNominalPct":40.11860700121606,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":32.24844964691052,"vramAllocatedGb":17.001568256,"vramAllocatedPct":67.30349582179929,"vramReservedGb":17.607688192,"vramReservedPct":69.7029209786926,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.31214848,"vramAllocatedPct":96.24362641165071,"vramReservedGb":24.354226176,"vramReservedPct":96.41019787107636,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 385.69 MiB is free. Including non-PyTorch memory, this process has 23.14 GiB memory in use. Of the allocated memory 22.64 GiB is allocated by PyTorch, and 40.13 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":39897.36939917021,"meanTps":39871.944338987996,"stepMs":102.66341018676758,"jitter":0.002058967962103397,"achievedTflops":64.69416283302131,"nominalPeakTflops":165.2,"mfuNominalPct":39.16111551635673,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":31.478791419846555,"vramAllocatedGb":10.234963968,"vramAllocatedPct":40.51678317460235,"vramReservedGb":10.554966016,"vramReservedPct":41.783563754854676,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":40802.44122627213,"meanTps":40801.6811841832,"stepMs":200.77230072021484,"jitter":0.0007183804740810469,"achievedTflops":66.16174992058828,"nominalPeakTflops":165.2,"mfuNominalPct":40.04948542408492,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":32.19288780500606,"vramAllocatedGb":17.001568256,"vramAllocatedPct":67.30349582179929,"vramReservedGb":17.607688192,"vramReservedPct":69.7029209786926,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.31214848,"vramAllocatedPct":96.24362641165071,"vramReservedGb":24.354226176,"vramReservedPct":96.41019787107636,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 385.69 MiB is free. Including non-PyTorch memory, this process has 23.14 GiB memory in use. Of the allocated memory 22.64 GiB is allocated by PyTorch, and 40.13 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":51990.2616353881,"meanTps":51985.59160971193,"stepMs":78.78398513793945,"jitter":0.0006679067469420173,"achievedTflops":84.30296289261447,"nominalPeakTflops":209.5,"mfuNominalPct":40.24007775303793,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":31.345588074939254,"vramAllocatedGb":10.234963968,"vramAllocatedPct":30.398904565680315,"vramReservedGb":10.575937536,"vramReservedPct":31.411631428760515,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":59220.42570089527,"meanTps":59220.23005597895,"stepMs":138.33065032958984,"jitter":0.0003028772328476624,"achievedTflops":96.02677873329264,"nominalPeakTflops":209.5,"mfuNominalPct":45.83617123307525,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":35.70474568220457,"vramAllocatedGb":17.001568256,"vramAllocatedPct":50.49642114001861,"vramReservedGb":17.607688192,"vramReservedPct":52.29666021730583,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":58000.55475691087,"meanTps":57973.573762017346,"stepMs":282.4800567626953,"jitter":0.0006902607103600124,"achievedTflops":94.04874031437272,"nominalPeakTflops":209.5,"mfuNominalPct":44.89200015005858,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":34.969270019802906,"vramAllocatedGb":30.534776832,"vramAllocatedPct":90.6914542886952,"vramReservedGb":31.717326848,"vramReservedPct":94.20375049148801,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":51787.37242786018,"meanTps":51775.32586505076,"stepMs":79.0926399230957,"jitter":0.00086530509976024,"achievedTflops":83.973975101526,"nominalPeakTflops":209.5,"mfuNominalPct":40.0830430078883,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":31.22326359870134,"vramAllocatedGb":10.234963968,"vramAllocatedPct":30.398904565680315,"vramReservedGb":10.575937536,"vramReservedPct":31.411631428760515,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":59016.25618246366,"meanTps":59009.62449220492,"stepMs":138.80921173095703,"jitter":0.00029425460622982267,"achievedTflops":95.69571489951444,"nominalPeakTflops":209.5,"mfuNominalPct":45.67814553676107,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":35.58164929028605,"vramAllocatedGb":17.001568256,"vramAllocatedPct":50.49642114001861,"vramReservedGb":17.607688192,"vramReservedPct":52.29666021730583,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":57869.4743595451,"meanTps":57861.27628726505,"stepMs":283.1199035644531,"jitter":0.0003153049177274399,"achievedTflops":93.8361915499027,"nominalPeakTflops":209.5,"mfuNominalPct":44.79054489255498,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":34.89023998588344,"vramAllocatedGb":30.534776832,"vramAllocatedPct":90.6914542886952,"vramReservedGb":31.717326848,"vramReservedPct":94.20375049148801,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":96828.61650760351,"meanTps":96819.60343055724,"stepMs":169.2061767578125,"jitter":0.0005440876965985325,"achievedTflops":157.009005294704,"nominalPeakTflops":989.5,"mfuNominalPct":15.867509377938758,"configuredPeakTflops":989.5,"mfuConfiguredPct":15.867509377938758,"vramAllocatedGb":30.584846336,"vramAllocatedPct":35.97476833936654,"vramReservedGb":31.744589824,"vramReservedPct":37.33889169822025,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":104083.13608681652,"meanTps":104075.514686175,"stepMs":314.82525634765625,"jitter":0.0002497061900929742,"achievedTflops":168.77231395391368,"nominalPeakTflops":989.5,"mfuNominalPct":17.056322784629984,"configuredPeakTflops":989.5,"mfuConfiguredPct":17.056322784629984,"vramAllocatedGb":57.651262976,"vramAllocatedPct":67.81105934778888,"vramReservedGb":60.070821888,"vramReservedPct":70.65701353001394,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.099464704,"vramAllocatedPct":98.92018834928743,"vramReservedGb":84.188069888,"vramReservedPct":99.0244082931462,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 40.19 MiB is free. Including non-PyTorch memory, this process has 79.13 GiB memory in use. Of the allocated memory 78.32 GiB is allocated by PyTorch, and 84.50 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":96794.08116025775,"meanTps":96751.39230706364,"stepMs":169.26654815673828,"jitter":0.0007187094354777461,"achievedTflops":156.95300572835845,"nominalPeakTflops":989.5,"mfuNominalPct":15.861849997812879,"configuredPeakTflops":989.5,"mfuConfiguredPct":15.861849997812879,"vramAllocatedGb":30.584846336,"vramAllocatedPct":35.97476833936654,"vramReservedGb":31.744589824,"vramReservedPct":37.33889169822025,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":103975.6398351309,"meanTps":103978.28740583459,"stepMs":315.15074157714844,"jitter":0.0007017572767403293,"achievedTflops":168.5980072235398,"nominalPeakTflops":989.5,"mfuNominalPct":17.038707147401702,"configuredPeakTflops":989.5,"mfuConfiguredPct":17.038707147401702,"vramAllocatedGb":57.651262976,"vramAllocatedPct":67.81105934778888,"vramReservedGb":60.070821888,"vramReservedPct":70.65701353001394,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.099464704,"vramAllocatedPct":98.92018834928743,"vramReservedGb":84.188069888,"vramReservedPct":99.0244082931462,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 40.19 MiB is free. Including non-PyTorch memory, this process has 79.13 GiB memory in use. Of the allocated memory 78.32 GiB is allocated by PyTorch, and 84.50 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":117751.16256512071,"meanTps":117776.40921492713,"stepMs":278.28175354003906,"jitter":0.0009104945624868094,"achievedTflops":190.93521702019578,"nominalPeakTflops":989.5,"mfuNominalPct":19.29613107834217,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.29613107834217,"vramAllocatedGb":57.651262976,"vramAllocatedPct":38.403258314412305,"vramReservedGb":59.99951872,"vramReservedPct":39.96750282996916,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":122999.77209787202,"meanTps":122893.83952727006,"stepMs":532.8139953613281,"jitter":0.0019513247906596668,"achievedTflops":199.4459134613958,"nominalPeakTflops":989.5,"mfuNominalPct":20.15623177982777,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.15623177982777,"vramAllocatedGb":111.784096768,"vramAllocatedPct":74.46278402282137,"vramReservedGb":116.480016384,"vramReservedPct":77.59087879000865,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.656285184,"vramAllocatedPct":99.02446928801338,"vramReservedGb":149.325611008,"vramReservedPct":99.47032756047273,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 449.44 MiB is free. Including non-PyTorch memory, this process has 139.36 GiB memory in use. Of the allocated memory 138.45 GiB is allocated by PyTorch, and 198.32 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":117865.06934635238,"meanTps":117782.86436647859,"stepMs":278.0128173828125,"jitter":0.0006003684992322658,"achievedTflops":191.1199185171556,"nominalPeakTflops":989.5,"mfuNominalPct":19.31479722255236,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.31479722255236,"vramAllocatedGb":57.651262976,"vramAllocatedPct":38.403258314412305,"vramReservedGb":59.99951872,"vramReservedPct":39.96750282996916,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":122841.4714592614,"meanTps":122869.46012287222,"stepMs":533.5006103515625,"jitter":0.001387890329960139,"achievedTflops":199.18922668116264,"nominalPeakTflops":989.5,"mfuNominalPct":20.13029072068344,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.13029072068344,"vramAllocatedGb":111.784096768,"vramAllocatedPct":74.46278402282137,"vramReservedGb":116.480016384,"vramReservedPct":77.59087879000865,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.656285184,"vramAllocatedPct":99.02446928801338,"vramReservedGb":149.325611008,"vramReservedPct":99.47032756047273,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 449.44 MiB is free. Including non-PyTorch memory, this process has 139.36 GiB memory in use. Of the allocated memory 138.45 GiB is allocated by PyTorch, and 198.32 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":36875.55638726762,"meanTps":36867.79917900048,"stepMs":111.07629013061523,"jitter":0.005746287632628949,"achievedTflops":59.79424923001984,"nominalPeakTflops":364.2,"mfuNominalPct":16.417970683695728,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":13.24541596220863,"vramAllocatedGb":10.234963968,"vramAllocatedPct":20.11713693082354,"vramReservedGb":10.554966016,"vramReservedPct":20.746110812694265,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":38310.58998401022,"meanTps":38252.653096860224,"stepMs":213.83121490478516,"jitter":0.0048701489138173746,"achievedTflops":62.121177009384986,"nominalPeakTflops":364.2,"mfuNominalPct":17.056885505048047,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":13.760868982333447,"vramAllocatedGb":17.001568256,"vramAllocatedPct":33.41710608010367,"vramReservedGb":17.607688192,"vramReservedPct":34.608453483683896,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":36684.931454431004,"meanTps":36690.00901390262,"stepMs":446.6138916015625,"jitter":0.0023491195822734055,"achievedTflops":59.48514813812624,"nominalPeakTflops":364.2,"mfuNominalPct":16.33309943386223,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":13.176944953888853,"vramAllocatedGb":30.534776832,"vramAllocatedPct":60.017044378663925,"vramReservedGb":31.696355328,"vramReservedPct":62.3001626908526,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":23287.528373588804,"meanTps":23294.821062021358,"stepMs":351.7762756347656,"jitter":0.0030890594911687818,"achievedTflops":37.76106483378555,"nominalPeakTflops":154.8,"mfuNominalPct":24.39345273500358,"configuredPeakTflops":180.6,"mfuConfiguredPct":20.908673772860215,"vramAllocatedGb":17.001568256,"vramAllocatedPct":33.31424341480078,"vramReservedGb":17.775460352,"vramReservedPct":34.83066997467626,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25201.96282713277,"meanTps":25193.747884119675,"stepMs":650.1080932617188,"jitter":0.001126352579400342,"achievedTflops":40.86534805184893,"nominalPeakTflops":154.8,"mfuNominalPct":26.3988036510652,"configuredPeakTflops":180.6,"mfuConfiguredPct":22.627545986627318,"vramAllocatedGb":30.534776832,"vramAllocatedPct":59.8323032723098,"vramReservedGb":31.702646784,"vramReservedPct":62.12072180358436,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.228967936,"vramAllocatedPct":96.4632083455329,"vramReservedGb":49.549410304,"vramReservedPct":97.09110895371354,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.25 GiB is free. Process 1415479 has 46.27 GiB memory in use. Of the allocated memory 45.85 GiB is allocated by PyTorch, and 105.60 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":23423.827271567898,"meanTps":23424.03494139123,"stepMs":349.72935485839844,"jitter":0.0011336265404378726,"achievedTflops":37.982075472647466,"nominalPeakTflops":154.8,"mfuNominalPct":24.536224465534538,"configuredPeakTflops":180.6,"mfuConfiguredPct":21.03104954188675,"vramAllocatedGb":17.001568256,"vramAllocatedPct":33.31424341480078,"vramReservedGb":17.775460352,"vramReservedPct":34.83066997467626,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25245.498002774944,"meanTps":25241.063785545986,"stepMs":648.9869995117188,"jitter":0.0006844805178743954,"achievedTflops":40.93594097023862,"nominalPeakTflops":154.8,"mfuNominalPct":26.444406311523654,"configuredPeakTflops":180.6,"mfuConfiguredPct":22.66663398130599,"vramAllocatedGb":30.534776832,"vramAllocatedPct":59.8323032723098,"vramReservedGb":31.702646784,"vramReservedPct":62.12072180358436,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.228967936,"vramAllocatedPct":96.4632083455329,"vramReservedGb":49.549410304,"vramReservedPct":97.09110895371354,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.25 GiB is free. Process 1449495 has 46.27 GiB memory in use. Of the allocated memory 45.85 GiB is allocated by PyTorch, and 105.60 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":70337.06952935315,"meanTps":70316.36820838375,"stepMs":116.46774673461914,"jitter":0.0003847406340524866,"achievedTflops":114.05257784800604,"nominalPeakTflops":468.0,"mfuNominalPct":24.370208941881632,"configuredPeakTflops":468.0,"mfuConfiguredPct":24.370208941881632,"vramAllocatedGb":17.001568256,"vramAllocatedPct":16.672462112764855,"vramReservedGb":17.607688192,"vramReservedPct":17.266849143219243,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":71914.77082036374,"meanTps":71913.63557677389,"stepMs":227.8252410888672,"jitter":0.0002664672416096842,"achievedTflops":116.61084336173762,"nominalPeakTflops":468.0,"mfuNominalPct":24.916846872166158,"configuredPeakTflops":468.0,"mfuConfiguredPct":24.916846872166158,"vramAllocatedGb":30.534776832,"vramAllocatedPct":29.943702968318103,"vramReservedGb":31.681675264,"vramReservedPct":31.06840042956326,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":64519.433010622546,"meanTps":64509.1541589924,"stepMs":507.87799072265625,"jitter":0.00026160417652867603,"achievedTflops":104.61919590042538,"nominalPeakTflops":468.0,"mfuNominalPct":22.354529038552432,"configuredPeakTflops":468.0,"mfuConfiguredPct":22.354529038552432,"vramAllocatedGb":57.601193472,"vramAllocatedPct":56.48618417733559,"vramReservedGb":59.942895616,"vramReservedPct":58.78255705819863,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":70043.64455171106,"meanTps":70032.12390646654,"stepMs":116.95565032958984,"jitter":0.0005019980807079839,"achievedTflops":113.5767849932709,"nominalPeakTflops":468.0,"mfuNominalPct":24.26854380198096,"configuredPeakTflops":468.0,"mfuConfiguredPct":24.26854380198096,"vramAllocatedGb":17.032730624,"vramAllocatedPct":16.70302125836842,"vramReservedGb":18.09842176,"vramReservedPct":17.748083385657704,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":71719.98860486777,"meanTps":71683.01521960333,"stepMs":228.44398498535156,"jitter":0.00039309737527890856,"achievedTflops":116.29500117574793,"nominalPeakTflops":468.0,"mfuNominalPct":24.849359225587165,"configuredPeakTflops":468.0,"mfuConfiguredPct":24.849359225587165,"vramAllocatedGb":30.56249856,"vramAllocatedPct":29.970888075763543,"vramReservedGb":33.948696576,"vramReservedPct":33.29153810510161,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":64407.25924375675,"meanTps":64406.022592899026,"stepMs":508.7625274658203,"jitter":0.0004909501068328869,"achievedTflops":104.43730451138157,"nominalPeakTflops":468.0,"mfuNominalPct":22.315663357132813,"configuredPeakTflops":468.0,"mfuConfiguredPct":22.315663357132813,"vramAllocatedGb":57.649526272,"vramAllocatedPct":56.53358138003302,"vramReservedGb":60.0834048,"vramReservedPct":58.92034634983698,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":45959.23830740991,"meanTps":45925.54307352165,"stepMs":356.48980712890625,"jitter":0.0022212744248192302,"achievedTflops":81.10174559808797,"nominalPeakTflops":312.0,"mfuNominalPct":25.994149230156403,"configuredPeakTflops":312.0,"mfuConfiguredPct":25.994149230156403,"vramAllocatedGb":30.541600768,"vramAllocatedPct":35.942187726553364,"vramReservedGb":31.704743936,"vramReservedPct":37.31100629027829,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":48624.65443400675,"meanTps":48617.54176100918,"stepMs":673.8968200683594,"jitter":0.001239469590305807,"achievedTflops":85.80525915865637,"nominalPeakTflops":312.0,"mfuNominalPct":27.501685627774478,"configuredPeakTflops":312.0,"mfuConfiguredPct":27.501685627774478,"vramAllocatedGb":57.61313792,"vramAllocatedPct":67.8007100664505,"vramReservedGb":60.005810176,"vramReservedPct":70.61647195288616,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.227264,"vramAllocatedPct":99.12093859709672,"vramReservedGb":84.374716416,"vramReservedPct":99.29446461679896,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 86.75 MiB is free. Process 764857 has 79.04 GiB memory in use. Of the allocated memory 78.42 GiB is allocated by PyTorch, and 124.63 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":45883.63010884727,"meanTps":45874.49693077519,"stepMs":357.0772399902344,"jitter":0.0033077721713739177,"achievedTflops":80.96832395946241,"nominalPeakTflops":312.0,"mfuNominalPct":25.95138588444308,"configuredPeakTflops":312.0,"mfuConfiguredPct":25.95138588444308,"vramAllocatedGb":30.541600768,"vramAllocatedPct":35.942187726553364,"vramReservedGb":31.704743936,"vramReservedPct":37.31100629027829,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":48589.72532737506,"meanTps":48575.433046553706,"stepMs":674.3812561035156,"jitter":0.0012083208869530786,"achievedTflops":85.74362168109279,"nominalPeakTflops":312.0,"mfuNominalPct":27.48193002599128,"configuredPeakTflops":312.0,"mfuConfiguredPct":27.48193002599128,"vramAllocatedGb":57.61313792,"vramAllocatedPct":67.8007100664505,"vramReservedGb":60.005810176,"vramReservedPct":70.61647195288616,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.227264,"vramAllocatedPct":99.12093859709672,"vramReservedGb":84.374716416,"vramReservedPct":99.29446461679896,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 86.75 MiB is free. Process 852916 has 79.04 GiB memory in use. Of the allocated memory 78.42 GiB is allocated by PyTorch, and 124.63 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":140672.46654328008,"meanTps":140617.7207696283,"stepMs":232.93826293945312,"jitter":0.001483440854425193,"achievedTflops":248.23698160396248,"nominalPeakTflops":2250.0,"mfuNominalPct":11.032754737953889,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.032754737953889,"vramAllocatedGb":57.61313792,"vramAllocatedPct":30.084716996725646,"vramReservedGb":60.045656064,"vramReservedPct":31.354941507899788,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":152060.87409575313,"meanTps":151993.01091532936,"stepMs":430.98529052734375,"jitter":0.0009077201132431975,"achievedTflops":268.33347941600516,"nominalPeakTflops":2250.0,"mfuNominalPct":11.925932418489118,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.925932418489118,"vramAllocatedGb":111.756211712,"vramAllocatedPct":58.35741852232159,"vramReservedGb":116.440170496,"vramReservedPct":60.80331158644593,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.520694272,"vramAllocatedPct":97.39831059016038,"vramReservedGb":187.273576448,"vramReservedPct":97.79145437670938,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.71 GiB is free. Including non-PyTorch memory, this process has 174.62 GiB memory in use. Of the allocated memory 173.71 GiB is allocated by PyTorch, and 98.00 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":140721.64035173797,"meanTps":140682.9925341866,"stepMs":232.85686492919922,"jitter":0.0014683663960716902,"achievedTflops":248.3237559250895,"nominalPeakTflops":2250.0,"mfuNominalPct":11.036611374448421,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.036611374448421,"vramAllocatedGb":57.61313792,"vramAllocatedPct":30.084716996725646,"vramReservedGb":60.045656064,"vramReservedPct":31.354941507899788,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":152017.23077994707,"meanTps":152019.21299631102,"stepMs":431.10902404785156,"jitter":0.00037089714669448845,"achievedTflops":268.2564644517474,"nominalPeakTflops":2250.0,"mfuNominalPct":11.922509531188773,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.922509531188773,"vramAllocatedGb":111.756211712,"vramAllocatedPct":58.35741852232159,"vramReservedGb":116.440170496,"vramReservedPct":60.80331158644593,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.520694272,"vramAllocatedPct":97.39831059016038,"vramReservedGb":187.273576448,"vramReservedPct":97.79145437670938,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.71 GiB is free. Including non-PyTorch memory, this process has 174.62 GiB memory in use. Of the allocated memory 173.71 GiB is allocated by PyTorch, and 98.00 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":153140.83351892218,"meanTps":152968.44621219143,"stepMs":427.9459533691406,"jitter":0.00018877522270127083,"achievedTflops":270.23922454189847,"nominalPeakTflops":2250.0,"mfuNominalPct":12.010632201862155,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.010632201862155,"vramAllocatedGb":111.756211712,"vramAllocatedPct":38.88137640472815,"vramReservedGb":116.415004672,"vramReservedPct":40.50222843518408,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":128,"tokensPerStep":131072,"status":"complete","stable":true,"tps":160823.18492616588,"meanTps":160824.16470410078,"stepMs":815.0068664550781,"jitter":5.318366990069139e-05,"achievedTflops":283.7958484628162,"nominalPeakTflops":2250.0,"mfuNominalPct":12.61314882056961,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.61314882056961,"vramAllocatedGb":220.042359808,"vramAllocatedPct":76.555474506665,"vramReservedGb":229.214519296,"vramReservedPct":79.74658290264541,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.046577152,"vramAllocatedPct":99.5191628738503,"vramReservedGb":286.181556224,"vramReservedPct":99.56612377226297,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 335.62 MiB is free. Including non-PyTorch memory, this process has 267.35 GiB memory in use. Of the allocated memory 266.40 GiB is allocated by PyTorch, and 128.73 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":153072.5601892483,"meanTps":152974.6796198752,"stepMs":428.13682556152344,"jitter":0.0002854936467275813,"achievedTflops":270.11874634386334,"nominalPeakTflops":2250.0,"mfuNominalPct":12.005277615282816,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.005277615282816,"vramAllocatedGb":111.756211712,"vramAllocatedPct":38.88137640472815,"vramReservedGb":116.415004672,"vramReservedPct":40.50222843518408,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":128,"tokensPerStep":131072,"status":"complete","stable":true,"tps":160828.44828648245,"meanTps":160828.67171592158,"stepMs":814.9801940917969,"jitter":0.00011258873950389794,"achievedTflops":283.80513642591364,"nominalPeakTflops":2250.0,"mfuNominalPct":12.613561618929497,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.613561618929497,"vramAllocatedGb":220.042359808,"vramAllocatedPct":76.555474506665,"vramReservedGb":229.214519296,"vramReservedPct":79.74658290264541,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.046577152,"vramAllocatedPct":99.5191628738503,"vramReservedGb":286.181556224,"vramReservedPct":99.56612377226297,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 335.62 MiB is free. Including non-PyTorch memory, this process has 267.35 GiB memory in use. Of the allocated memory 266.40 GiB is allocated by PyTorch, and 128.73 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":38797.31434905614,"meanTps":38794.497103113805,"stepMs":105.5743179321289,"jitter":0.0012059257743822244,"achievedTflops":68.4634914351679,"nominalPeakTflops":165.2,"mfuNominalPct":41.44279142564643,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":33.31286583496938,"vramAllocatedGb":10.237947904,"vramAllocatedPct":40.52859557455773,"vramReservedGb":10.552868864,"vramReservedPct":41.775261834776224,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":39901.9404066532,"meanTps":39882.67262244522,"stepMs":205.3032989501953,"jitter":0.0006164113108843214,"achievedTflops":70.41276441713138,"nominalPeakTflops":165.2,"mfuNominalPct":42.622738751290186,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":34.26134024027154,"vramAllocatedGb":17.005832192,"vramAllocatedPct":67.3203753116463,"vramReservedGb":17.609785344,"vramReservedPct":69.71122289877106,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.313852416,"vramAllocatedPct":96.25037172171444,"vramReservedGb":24.343740416,"vramReservedPct":96.3686882706841,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 395.69 MiB is free. Including non-PyTorch memory, this process has 23.13 GiB memory in use. Of the allocated memory 22.64 GiB is allocated by PyTorch, and 28.50 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":38674.06063908368,"meanTps":38651.76366181782,"stepMs":105.91078186035156,"jitter":0.002304491022795232,"achievedTflops":68.2459923773431,"nominalPeakTflops":165.2,"mfuNominalPct":41.31113340032876,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":33.207035460551346,"vramAllocatedGb":10.237947904,"vramAllocatedPct":40.52859557455773,"vramReservedGb":10.552868864,"vramReservedPct":41.775261834776224,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":39841.952828994996,"meanTps":39833.87313913462,"stepMs":205.61241149902344,"jitter":0.0007874000066048164,"achievedTflops":70.30690763095622,"nominalPeakTflops":165.2,"mfuNominalPct":42.55866079355704,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":34.209832599605654,"vramAllocatedGb":17.005832192,"vramAllocatedPct":67.3203753116463,"vramReservedGb":17.609785344,"vramReservedPct":69.71122289877106,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.313852416,"vramAllocatedPct":96.25037172171444,"vramReservedGb":24.343740416,"vramReservedPct":96.3686882706841,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 395.69 MiB is free. Including non-PyTorch memory, this process has 23.13 GiB memory in use. Of the allocated memory 22.64 GiB is allocated by PyTorch, and 28.50 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":50370.033113359445,"meanTps":50397.91913750361,"stepMs":81.31819152832031,"jitter":0.0016318487488621797,"achievedTflops":88.88523312772816,"nominalPeakTflops":209.5,"mfuNominalPct":42.42731891538337,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":33.04937107745235,"vramAllocatedGb":10.237947904,"vramAllocatedPct":30.407767165097148,"vramReservedGb":10.573840384,"vramReservedPct":31.40540266980181,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":57551.02068818842,"meanTps":57549.986422923575,"stepMs":142.34326171875,"jitter":0.000343130872258755,"achievedTflops":101.55712780843074,"nominalPeakTflops":209.5,"mfuNominalPct":48.47595599447768,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":37.761044038416756,"vramAllocatedGb":17.005832192,"vramAllocatedPct":50.50908547219832,"vramReservedGb":17.609785344,"vramReservedPct":52.30288897626453,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":56570.50392782734,"meanTps":56572.644560053195,"stepMs":289.6208953857422,"jitter":0.0003203346681176222,"achievedTflops":99.82686369218135,"nominalPeakTflops":209.5,"mfuNominalPct":47.65005426834431,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":37.117695994791184,"vramAllocatedGb":30.541600768,"vramAllocatedPct":90.71172208640067,"vramReservedGb":31.704743936,"vramReservedPct":94.16637793773577,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":50245.90023088722,"meanTps":50212.42996131053,"stepMs":81.51908874511719,"jitter":0.0007393664218656848,"achievedTflops":88.66618264244201,"nominalPeakTflops":209.5,"mfuNominalPct":42.322760211189504,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":32.967923569039854,"vramAllocatedGb":10.237947904,"vramAllocatedPct":30.407767165097148,"vramReservedGb":10.573840384,"vramReservedPct":31.40540266980181,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":57480.785693243364,"meanTps":57468.36755062118,"stepMs":142.5171890258789,"jitter":0.0005683560605659107,"achievedTflops":101.43318796734772,"nominalPeakTflops":209.5,"mfuNominalPct":48.41679616579843,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":37.714960637888936,"vramAllocatedGb":17.005832192,"vramAllocatedPct":50.50908547219832,"vramReservedGb":17.609785344,"vramReservedPct":52.30288897626453,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":56557.39005686386,"meanTps":56559.689684657686,"stepMs":289.68804931640625,"jitter":0.00025958195619319596,"achievedTflops":99.80372236378133,"nominalPeakTflops":209.5,"mfuNominalPct":47.63900828820111,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":37.109091569482224,"vramAllocatedGb":30.541600768,"vramAllocatedPct":90.71172208640067,"vramReservedGb":31.704743936,"vramReservedPct":94.16637793773577,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":94050.45425948883,"meanTps":94038.73016015647,"stepMs":174.2043685913086,"jitter":0.0002788805304880172,"achievedTflops":165.9656751427902,"nominalPeakTflops":989.5,"mfuNominalPct":16.772680661221848,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.772680661221848,"vramAllocatedGb":30.591670272,"vramAllocatedPct":35.98279484746359,"vramReservedGb":31.746686976,"vramReservedPct":37.341358428199655,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":101138.55079103666,"meanTps":101123.9373394301,"stepMs":323.99119567871094,"jitter":0.0002252554044543782,"achievedTflops":178.47365009727497,"nominalPeakTflops":989.5,"mfuNominalPct":18.036750894115713,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.036750894115713,"vramAllocatedGb":57.663207424,"vramAllocatedPct":67.8251087481037,"vramReservedGb":60.068724736,"vramReservedPct":70.65454680003454,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.159892992,"vramAllocatedPct":98.99126582465095,"vramReservedGb":84.227915776,"vramReservedPct":99.07127616275484,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 2.19 MiB is free. Including non-PyTorch memory, this process has 79.17 GiB memory in use. Of the allocated memory 78.38 GiB is allocated by PyTorch, and 64.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":94005.2123277645,"meanTps":93999.46067754415,"stepMs":174.2882080078125,"jitter":0.000908732662636603,"achievedTflops":165.88583918875347,"nominalPeakTflops":989.5,"mfuNominalPct":16.764612348534964,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.764612348534964,"vramAllocatedGb":30.591670272,"vramAllocatedPct":35.98279484746359,"vramReservedGb":31.746686976,"vramReservedPct":37.341358428199655,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":101060.1655865186,"meanTps":101045.38269478563,"stepMs":324.24249267578125,"jitter":0.0006995925091328967,"achievedTflops":178.335328028642,"nominalPeakTflops":989.5,"mfuNominalPct":18.02277190789712,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.02277190789712,"vramAllocatedGb":57.663207424,"vramAllocatedPct":67.8251087481037,"vramReservedGb":60.068724736,"vramReservedPct":70.65454680003454,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.159892992,"vramAllocatedPct":98.99126582465095,"vramReservedGb":84.227915776,"vramReservedPct":99.07127616275484,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 2.19 MiB is free. Including non-PyTorch memory, this process has 79.17 GiB memory in use. Of the allocated memory 78.38 GiB is allocated by PyTorch, and 64.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":114184.97351526501,"meanTps":114216.2497320751,"stepMs":286.97296142578125,"jitter":0.0013701018660013846,"achievedTflops":201.49595629104164,"nominalPeakTflops":989.5,"mfuNominalPct":20.363411449322044,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.363411449322044,"vramAllocatedGb":57.663207424,"vramAllocatedPct":38.4112148742219,"vramReservedGb":60.003713024,"vramReservedPct":39.97029678333023,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":118892.33509812056,"meanTps":118853.40628790832,"stepMs":551.2214050292969,"jitter":0.0024169868904077817,"achievedTflops":209.80277893630318,"nominalPeakTflops":989.5,"mfuNominalPct":21.202908432168083,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.202908432168083,"vramAllocatedGb":111.806281216,"vramAllocatedPct":74.47756175782885,"vramReservedGb":116.503085056,"vramReservedPct":77.60624553349449,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.65798912,"vramAllocatedPct":99.0256043315663,"vramReservedGb":149.329805312,"vramReservedPct":99.47312151383377,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 445.44 MiB is free. Including non-PyTorch memory, this process has 139.37 GiB memory in use. Of the allocated memory 138.45 GiB is allocated by PyTorch, and 200.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":114043.78209978531,"meanTps":114013.06452605114,"stepMs":287.3282470703125,"jitter":0.0012900667667037839,"achievedTflops":201.24680354872953,"nominalPeakTflops":989.5,"mfuNominalPct":20.338231788653815,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.338231788653815,"vramAllocatedGb":57.663207424,"vramAllocatedPct":38.4112148742219,"vramReservedGb":60.003713024,"vramReservedPct":39.97029678333023,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":118718.74372886612,"meanTps":118748.60019356031,"stepMs":552.0274047851562,"jitter":0.0018080034014353826,"achievedTflops":209.49645177367418,"nominalPeakTflops":989.5,"mfuNominalPct":21.171950659289966,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.171950659289966,"vramAllocatedGb":111.806281216,"vramAllocatedPct":74.47756175782885,"vramReservedGb":116.503085056,"vramReservedPct":77.60624553349449,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.65798912,"vramAllocatedPct":99.0256043315663,"vramReservedGb":149.329805312,"vramReservedPct":99.47312151383377,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 445.44 MiB is free. Including non-PyTorch memory, this process has 139.37 GiB memory in use. Of the allocated memory 138.45 GiB is allocated by PyTorch, and 200.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":36128.33055694041,"meanTps":36102.77175470991,"stepMs":113.37363052368164,"jitter":0.004115631632164731,"achievedTflops":63.75368221105178,"nominalPeakTflops":364.2,"mfuNominalPct":17.505129657070782,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.122495906912176,"vramAllocatedGb":10.237947904,"vramAllocatedPct":20.12300194894109,"vramReservedGb":10.552868864,"vramReservedPct":20.741988795842946,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":37342.08391524828,"meanTps":37349.51401851071,"stepMs":219.3771514892578,"jitter":0.00297858507080113,"achievedTflops":65.89552615167338,"nominalPeakTflops":364.2,"mfuNominalPct":18.093225192661556,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.59694979311343,"vramAllocatedGb":17.005832192,"vramAllocatedPct":33.425486977647076,"vramReservedGb":17.609785344,"vramReservedPct":34.612575500535215,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":35747.3212609507,"meanTps":35662.358685146275,"stepMs":458.3280487060547,"jitter":0.001783987581710509,"achievedTflops":63.08133601620891,"nominalPeakTflops":364.2,"mfuNominalPct":17.320520597531278,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":13.973560095592864,"vramAllocatedGb":30.541600768,"vramAllocatedPct":60.030457035059044,"vramReservedGb":31.683772416,"vramReservedPct":62.27543058974468,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":22884.690885201715,"meanTps":22889.749400795627,"stepMs":357.96856689453125,"jitter":0.0016433850410156458,"achievedTflops":40.38335809327969,"nominalPeakTflops":154.8,"mfuNominalPct":26.08744062873365,"configuredPeakTflops":180.6,"mfuConfiguredPct":22.360663396057415,"vramAllocatedGb":17.005832192,"vramAllocatedPct":33.32259851473451,"vramReservedGb":17.777557504,"vramReservedPct":34.834779303366055,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":24686.01731030051,"meanTps":24682.278987306392,"stepMs":663.695556640625,"jitter":0.0005425798428322709,"achievedTflops":43.56205997885732,"nominalPeakTflops":154.8,"mfuNominalPct":28.140865619416875,"configuredPeakTflops":180.6,"mfuConfiguredPct":24.12074195950018,"vramAllocatedGb":30.541600768,"vramAllocatedPct":59.84567464261682,"vramReservedGb":31.706841088,"vramReservedPct":62.128940460963946,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.232720384,"vramAllocatedPct":96.47056119464605,"vramReservedGb":49.551507456,"vramReservedPct":97.09521828240334,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.25 GiB is free. Process 1415479 has 46.28 GiB memory in use. Of the allocated memory 45.85 GiB is allocated by PyTorch, and 104.02 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":22999.36503062307,"meanTps":22994.488313562757,"stepMs":356.1837463378906,"jitter":0.0009144121169738513,"achievedTflops":40.585717264387654,"nominalPeakTflops":154.8,"mfuNominalPct":26.21816360748556,"configuredPeakTflops":180.6,"mfuConfiguredPct":22.472711663559057,"vramAllocatedGb":17.005832192,"vramAllocatedPct":33.32259851473451,"vramReservedGb":17.777557504,"vramReservedPct":34.834779303366055,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":24730.45097953139,"meanTps":24728.828089391587,"stepMs":662.5030822753906,"jitter":0.0008756046471445054,"achievedTflops":43.64046963642929,"nominalPeakTflops":154.8,"mfuNominalPct":28.191517852990494,"configuredPeakTflops":180.6,"mfuConfiguredPct":24.16415815970614,"vramAllocatedGb":30.541600768,"vramAllocatedPct":59.84567464261682,"vramReservedGb":31.706841088,"vramReservedPct":62.128940460963946,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.232720384,"vramAllocatedPct":96.47056119464605,"vramReservedGb":49.551507456,"vramReservedPct":97.09521828240334,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.25 GiB is free. Process 1449495 has 46.28 GiB memory in use. Of the allocated memory 45.85 GiB is allocated by PyTorch, and 104.02 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":68790.658006954,"meanTps":68784.23164238015,"stepMs":119.0859375,"jitter":0.0014483739782707634,"achievedTflops":121.39109895356032,"nominalPeakTflops":468.0,"mfuNominalPct":25.93826900717101,"configuredPeakTflops":468.0,"mfuConfiguredPct":25.93826900717101,"vramAllocatedGb":17.005832192,"vramAllocatedPct":16.67664351005367,"vramReservedGb":17.609785344,"vramReservedPct":17.268905699810862,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":70738.4424281044,"meanTps":70697.60755959949,"stepMs":231.61380767822266,"jitter":0.0004790916818369801,"achievedTflops":124.82824722715539,"nominalPeakTflops":468.0,"mfuNominalPct":26.672702398964827,"configuredPeakTflops":468.0,"mfuConfiguredPct":26.672702398964827,"vramAllocatedGb":30.541600768,"vramAllocatedPct":29.950394810665045,"vramReservedGb":31.706841088,"vramReservedPct":31.093079108662668,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":63772.57389612324,"meanTps":63781.56056677472,"stepMs":513.8258972167969,"jitter":0.00026663302656338724,"achievedTflops":112.53596131563326,"nominalPeakTflops":468.0,"mfuNominalPct":24.04614558026352,"configuredPeakTflops":468.0,"mfuConfiguredPct":24.04614558026352,"vramAllocatedGb":57.61313792,"vramAllocatedPct":56.4978974118878,"vramReservedGb":59.942895616,"vramReservedPct":58.78255705819863,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":68394.06107631382,"meanTps":68376.02059311018,"stepMs":119.77648162841797,"jitter":0.0013330737216200703,"achievedTflops":120.69124611529914,"nominalPeakTflops":468.0,"mfuNominalPct":25.788727802414346,"configuredPeakTflops":468.0,"mfuConfiguredPct":25.788727802414346,"vramAllocatedGb":17.04006656,"vramAllocatedPct":16.710215189726984,"vramReservedGb":18.100518912,"vramReservedPct":17.75013994224932,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":70558.65757542245,"meanTps":70524.86197325561,"stepMs":232.20396423339844,"jitter":0.0005527458550128308,"achievedTflops":124.51099076422025,"nominalPeakTflops":468.0,"mfuNominalPct":26.60491255645732,"configuredPeakTflops":468.0,"mfuConfiguredPct":26.60491255645732,"vramAllocatedGb":30.570346496,"vramAllocatedPct":29.978584096133737,"vramReservedGb":33.961279488,"vramReservedPct":33.30387744465131,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":63699.74142602368,"meanTps":63679.024953341526,"stepMs":514.4133911132812,"jitter":0.0007970505267526712,"achievedTflops":112.40743785269453,"nominalPeakTflops":468.0,"mfuNominalPct":24.018683301857806,"configuredPeakTflops":468.0,"mfuConfiguredPct":24.018683301857806,"vramAllocatedGb":57.66147072,"vramAllocatedPct":56.54529461458523,"vramReservedGb":60.100182016,"vramReservedPct":58.936798802569925,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":24434.48942326156,"meanTps":24297.568059109515,"stepMs":83.81595230102539,"jitter":0.02189387173560131,"achievedTflops":50.112849956995774,"nominalPeakTflops":312.0,"mfuNominalPct":16.061810883652495,"configuredPeakTflops":312.0,"mfuConfiguredPct":16.061810883652495,"vramAllocatedGb":6.857717248,"vramAllocatedPct":8.070348459321428,"vramReservedGb":6.887047168,"vramReservedPct":8.104864708114428,"warmupSteps":22,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":33729.68406000741,"meanTps":33659.03228130003,"stepMs":121.43606185913086,"jitter":0.005440033210052166,"achievedTflops":69.17642382929729,"nominalPeakTflops":312.0,"mfuNominalPct":22.171930714518364,"configuredPeakTflops":312.0,"mfuConfiguredPct":22.171930714518364,"vramAllocatedGb":10.241995776,"vramAllocatedPct":12.053059617662756,"vramReservedGb":10.578034688,"vramReservedPct":12.448519362889517,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":39769.74904847257,"meanTps":39753.88957979217,"stepMs":205.98571014404297,"jitter":0.0035632095593018977,"achievedTflops":81.56403157727435,"nominalPeakTflops":312.0,"mfuNominalPct":26.142317813228956,"configuredPeakTflops":312.0,"mfuConfiguredPct":26.142317813228956,"vramAllocatedGb":17.010520064,"vramAllocatedPct":20.018443372070426,"vramReservedGb":17.613979648,"vramReservedPct":20.72861104855453,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":43615.62599633777,"meanTps":43632.90491014215,"stepMs":375.6451873779297,"jitter":0.0034610464913414148,"achievedTflops":89.45156510019547,"nominalPeakTflops":312.0,"mfuNominalPct":28.670373429549834,"configuredPeakTflops":312.0,"mfuConfiguredPct":28.670373429549834,"vramAllocatedGb":30.54756864,"vramAllocatedPct":35.94921088088576,"vramReservedGb":31.72990976,"vramReservedPct":37.34062211746994,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":46113.18303377538,"meanTps":46095.37572692317,"stepMs":710.5993957519531,"jitter":0.0020696618640020296,"achievedTflops":94.57382073271948,"nominalPeakTflops":312.0,"mfuNominalPct":30.312122029717784,"configuredPeakTflops":312.0,"mfuConfiguredPct":30.312122029717784,"vramAllocatedGb":57.621665792,"vramAllocatedPct":67.81074589851643,"vramReservedGb":60.049850368,"vramReservedPct":70.66829965047154,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.230680064,"vramAllocatedPct":99.12495871426434,"vramReservedGb":84.397785088,"vramReservedPct":99.3216124583913,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 44.75 MiB is free. Process 703773 has 79.09 GiB memory in use. Of the allocated memory 78.42 GiB is allocated by PyTorch, and 163.38 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":23562.784793670944,"meanTps":23441.144981168003,"stepMs":86.91672134399414,"jitter":0.026973527916566105,"achievedTflops":48.3250653811533,"nominalPeakTflops":312.0,"mfuNominalPct":15.488803006779904,"configuredPeakTflops":312.0,"mfuConfiguredPct":15.488803006779904,"vramAllocatedGb":6.857717248,"vramAllocatedPct":8.070348459321428,"vramReservedGb":6.887047168,"vramReservedPct":8.104864708114428,"warmupSteps":18,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":33599.21870856293,"meanTps":33225.36559111247,"stepMs":121.90759658813477,"jitter":0.00855321574429916,"achievedTflops":68.90885160921644,"nominalPeakTflops":312.0,"mfuNominalPct":22.086170387569375,"configuredPeakTflops":312.0,"mfuConfiguredPct":22.086170387569375,"vramAllocatedGb":10.241995776,"vramAllocatedPct":12.053059617662756,"vramReservedGb":10.578034688,"vramReservedPct":12.448519362889517,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":39703.10114733684,"meanTps":39680.46353431501,"stepMs":206.33148956298828,"jitter":0.004606461166410746,"achievedTflops":81.42734297242123,"nominalPeakTflops":312.0,"mfuNominalPct":26.098507362955523,"configuredPeakTflops":312.0,"mfuConfiguredPct":26.098507362955523,"vramAllocatedGb":17.010520064,"vramAllocatedPct":20.018443372070426,"vramReservedGb":17.613979648,"vramReservedPct":20.72861104855453,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":43644.77713862053,"meanTps":43630.56935658932,"stepMs":375.394287109375,"jitter":0.005009065557698123,"achievedTflops":89.51135136353766,"nominalPeakTflops":312.0,"mfuNominalPct":28.689535693441556,"configuredPeakTflops":312.0,"mfuConfiguredPct":28.689535693441556,"vramAllocatedGb":30.54756864,"vramAllocatedPct":35.94921088088576,"vramReservedGb":31.72990976,"vramReservedPct":37.34062211746994,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":46153.610443368176,"meanTps":46143.10411939075,"stepMs":709.9769592285156,"jitter":0.001525837806703554,"achievedTflops":94.6567335644951,"nominalPeakTflops":312.0,"mfuNominalPct":30.338696655286892,"configuredPeakTflops":312.0,"mfuConfiguredPct":30.338696655286892,"vramAllocatedGb":57.621665792,"vramAllocatedPct":67.81074589851643,"vramReservedGb":60.049850368,"vramReservedPct":70.66829965047154,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.230680064,"vramAllocatedPct":99.12495871426434,"vramReservedGb":84.397785088,"vramReservedPct":99.3216124583913,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 44.75 MiB is free. Process 736416 has 79.09 GiB memory in use. Of the allocated memory 78.42 GiB is allocated by PyTorch, and 163.38 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":12519.283913194056,"meanTps":12833.687548595673,"stepMs":163.58763122558594,"jitter":0.02432364058021735,"achievedTflops":25.675879100369652,"nominalPeakTflops":2250.0,"mfuNominalPct":1.1411501822386512,"configuredPeakTflops":2250.0,"mfuConfiguredPct":1.1411501822386512,"vramAllocatedGb":6.857717248,"vramAllocatedPct":3.580997149228775,"vramReservedGb":6.887047168,"vramReservedPct":3.596312793795156,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":25524.734874729234,"meanTps":25233.48830859135,"stepMs":160.47179412841797,"jitter":0.0065440380397634055,"achievedTflops":52.348841296093866,"nominalPeakTflops":2250.0,"mfuNominalPct":2.326615168715283,"configuredPeakTflops":2250.0,"mfuConfiguredPct":2.326615168715283,"vramAllocatedGb":10.241995776,"vramAllocatedPct":5.3482166659708215,"vramReservedGb":10.578034688,"vramReservedPct":5.523691148569661,"warmupSteps":13,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":49841.68554050476,"meanTps":49777.31344868082,"stepMs":164.36041259765625,"jitter":0.0333939025619773,"achievedTflops":102.22063026687474,"nominalPeakTflops":2250.0,"mfuNominalPct":4.543139122972211,"configuredPeakTflops":2250.0,"mfuConfiguredPct":4.543139122972211,"vramAllocatedGb":17.010520064,"vramAllocatedPct":8.882638588496508,"vramReservedGb":17.613979648,"vramReservedPct":9.19775613735856,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":103924.87928289303,"meanTps":103465.23030362462,"stepMs":157.65233612060547,"jitter":0.01689823912892721,"achievedTflops":213.14019671491658,"nominalPeakTflops":2250.0,"mfuNominalPct":9.472897631774071,"configuredPeakTflops":2250.0,"mfuConfiguredPct":9.472897631774071,"vramAllocatedGb":30.54756864,"vramAllocatedPct":15.951482433547882,"vramReservedGb":31.72990976,"vramReservedPct":16.568883243033103,"warmupSteps":17,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":130679.80902772813,"meanTps":130673.25634858335,"stepMs":250.7502899169922,"jitter":0.0008523981646634116,"achievedTflops":268.0120525039915,"nominalPeakTflops":2250.0,"mfuNominalPct":11.911646777955179,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.911646777955179,"vramAllocatedGb":57.621665792,"vramAllocatedPct":30.08917012365063,"vramReservedGb":60.049850368,"vramReservedPct":31.35713171057567,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":140245.96410185375,"meanTps":140234.6544914067,"stepMs":467.2933044433594,"jitter":0.0007517441016408609,"achievedTflops":287.6313408627913,"nominalPeakTflops":2250.0,"mfuNominalPct":12.783615149457393,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.783615149457393,"vramAllocatedGb":111.769860096,"vramAllocatedPct":58.36454550385613,"vramReservedGb":116.505182208,"vramReservedPct":60.837259727922074,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.528198656,"vramAllocatedPct":97.40222926699391,"vramReservedGb":187.63218944,"vramReservedPct":97.97871670549713,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.61 GiB is free. Including non-PyTorch memory, this process has 174.73 GiB memory in use. Of the allocated memory 173.72 GiB is allocated by PyTorch, and 192.85 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":12567.112390669094,"meanTps":12558.301359469426,"stepMs":162.9650421142578,"jitter":0.02257797502770813,"achievedTflops":25.773970829394948,"nominalPeakTflops":2250.0,"mfuNominalPct":1.1455098146397755,"configuredPeakTflops":2250.0,"mfuConfiguredPct":1.1455098146397755,"vramAllocatedGb":6.857717248,"vramAllocatedPct":3.580997149228775,"vramReservedGb":6.887047168,"vramReservedPct":3.596312793795156,"warmupSteps":35,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":24769.138887380497,"meanTps":24734.670898384324,"stepMs":165.36707305908203,"jitter":0.020262069740922175,"achievedTflops":50.79918467400509,"nominalPeakTflops":2250.0,"mfuNominalPct":2.257741541066893,"configuredPeakTflops":2250.0,"mfuConfiguredPct":2.257741541066893,"vramAllocatedGb":10.241995776,"vramAllocatedPct":5.3482166659708215,"vramReservedGb":10.578034688,"vramReservedPct":5.523691148569661,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":49592.35462274468,"meanTps":49032.735517255605,"stepMs":165.18675231933594,"jitter":0.04555699530717657,"achievedTflops":101.7092758999013,"nominalPeakTflops":2250.0,"mfuNominalPct":4.520412262217835,"configuredPeakTflops":2250.0,"mfuConfiguredPct":4.520412262217835,"vramAllocatedGb":17.010520064,"vramAllocatedPct":8.882638588496508,"vramReservedGb":17.613979648,"vramReservedPct":9.19775613735856,"warmupSteps":27,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":96312.18391367528,"meanTps":95321.71309820816,"stepMs":170.11347198486328,"jitter":0.023114665454671196,"achievedTflops":197.52727130454377,"nominalPeakTflops":2250.0,"mfuNominalPct":8.7789898357575,"configuredPeakTflops":2250.0,"mfuConfiguredPct":8.7789898357575,"vramAllocatedGb":30.54756864,"vramAllocatedPct":15.951482433547882,"vramReservedGb":31.72990976,"vramReservedPct":16.568883243033103,"warmupSteps":17,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":128940.36797309421,"meanTps":128473.83852798046,"stepMs":254.1329803466797,"jitter":0.003600153594815783,"achievedTflops":264.44462176828216,"nominalPeakTflops":2250.0,"mfuNominalPct":11.75309430081254,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.75309430081254,"vramAllocatedGb":57.621665792,"vramAllocatedPct":30.08917012365063,"vramReservedGb":60.049850368,"vramReservedPct":31.35713171057567,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":139865.17792972559,"meanTps":139846.10136821357,"stepMs":468.5655212402344,"jitter":0.002442811301713467,"achievedTflops":286.85038407752734,"nominalPeakTflops":2250.0,"mfuNominalPct":12.748905959001215,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.748905959001215,"vramAllocatedGb":111.769860096,"vramAllocatedPct":58.36454550385613,"vramReservedGb":116.505182208,"vramReservedPct":60.837259727922074,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.528198656,"vramAllocatedPct":97.40222926699391,"vramReservedGb":187.63218944,"vramReservedPct":97.97871670549713,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.61 GiB is free. Including non-PyTorch memory, this process has 174.73 GiB memory in use. Of the allocated memory 173.72 GiB is allocated by PyTorch, and 192.85 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":27705.598366782855,"meanTps":27365.094751664386,"stepMs":73.92007827758789,"jitter":0.003727367312073334,"achievedTflops":56.82158811968536,"nominalPeakTflops":2250.0,"mfuNominalPct":2.5254039164304602,"configuredPeakTflops":2250.0,"mfuConfiguredPct":2.5254039164304602,"vramAllocatedGb":6.857717248,"vramAllocatedPct":2.3858851468929476,"vramReservedGb":6.887047168,"vramReservedPct":2.3960893909521452,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":54653.429416193605,"meanTps":54309.510779236516,"stepMs":74.94497680664062,"jitter":0.010477840646030596,"achievedTflops":112.08906642270999,"nominalPeakTflops":2250.0,"mfuNominalPct":4.981736285453777,"configuredPeakTflops":2250.0,"mfuConfiguredPct":4.981736285453777,"vramAllocatedGb":10.241995776,"vramAllocatedPct":3.5633177503236,"vramReservedGb":10.578034688,"vramReservedPct":3.680229868441723,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":91923.91030811974,"meanTps":91933.34663034278,"stepMs":89.11718368530273,"jitter":0.0013239712715795429,"achievedTflops":188.52733302238363,"nominalPeakTflops":2250.0,"mfuNominalPct":8.378992578772607,"configuredPeakTflops":2250.0,"mfuConfiguredPct":8.378992578772607,"vramAllocatedGb":17.010520064,"vramAllocatedPct":5.918171556790041,"vramReservedGb":17.613979648,"vramReservedPct":6.128122653656232,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":114842.91684940513,"meanTps":114785.17901084077,"stepMs":142.66443634033203,"jitter":0.0015353798826726948,"achievedTflops":235.53206948613948,"nominalPeakTflops":2250.0,"mfuNominalPct":10.468091977161755,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.468091977161755,"vramAllocatedGb":30.54756864,"vramAllocatedPct":10.627879169722922,"vramReservedGb":31.72990976,"vramReservedPct":11.039230354782568,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":130576.2025414974,"meanTps":130296.2901345208,"stepMs":250.94924926757812,"jitter":0.00040940909182115513,"achievedTflops":267.7995653016148,"nominalPeakTflops":2250.0,"mfuNominalPct":11.90220290229399,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.90220290229399,"vramAllocatedGb":57.621665792,"vramAllocatedPct":20.047294395588686,"vramReservedGb":60.049850368,"vramReservedPct":20.89209001842988,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":141615.41175167597,"meanTps":141574.3012081558,"stepMs":462.7744903564453,"jitter":0.0003732361451839727,"achievedTflops":290.43994976845426,"nominalPeakTflops":2250.0,"mfuNominalPct":12.9084422119313,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.9084422119313,"vramAllocatedGb":111.769860096,"vramAllocatedPct":38.88612484732021,"vramReservedGb":116.505182208,"vramReservedPct":40.53360232185002,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":64,"tokensPerStep":131072,"status":"complete","stable":true,"tps":148584.5080218775,"meanTps":148564.27861462324,"stepMs":882.1377258300781,"jitter":6.621614716641304e-05,"achievedTflops":304.7329136882156,"nominalPeakTflops":2250.0,"mfuNominalPct":13.543685052809582,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.543685052809582,"vramAllocatedGb":220.066248192,"vramAllocatedPct":76.5637855726521,"vramReservedGb":229.55425792,"vramReservedPct":79.86478219659617,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":28103.58530408036,"meanTps":28096.346519266994,"stepMs":72.87326431274414,"jitter":0.0019026787272076553,"achievedTflops":57.637822063769626,"nominalPeakTflops":2250.0,"mfuNominalPct":2.5616809806119836,"configuredPeakTflops":2250.0,"mfuConfiguredPct":2.5616809806119836,"vramAllocatedGb":6.857717248,"vramAllocatedPct":2.3858851468929476,"vramReservedGb":6.887047168,"vramReservedPct":2.3960893909521452,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":55218.17240162644,"meanTps":55229.4730969903,"stepMs":74.1784782409668,"jitter":0.008384257000470494,"achievedTflops":113.24730140781752,"nominalPeakTflops":2250.0,"mfuNominalPct":5.033213395903001,"configuredPeakTflops":2250.0,"mfuConfiguredPct":5.033213395903001,"vramAllocatedGb":10.241995776,"vramAllocatedPct":3.5633177503236,"vramReservedGb":10.578034688,"vramReservedPct":3.680229868441723,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":92015.05279176224,"meanTps":92002.8751727916,"stepMs":89.02891159057617,"jitter":0.0006799846519008698,"achievedTflops":188.71425772248136,"nominalPeakTflops":2250.0,"mfuNominalPct":8.387300343221394,"configuredPeakTflops":2250.0,"mfuConfiguredPct":8.387300343221394,"vramAllocatedGb":17.010520064,"vramAllocatedPct":5.918171556790041,"vramReservedGb":17.613979648,"vramReservedPct":6.128122653656232,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":114910.1940164375,"meanTps":114783.79917951327,"stepMs":142.5809097290039,"jitter":0.0005965825260797293,"achievedTflops":235.67004865642716,"nominalPeakTflops":2250.0,"mfuNominalPct":10.474224384730096,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.474224384730096,"vramAllocatedGb":30.54756864,"vramAllocatedPct":10.627879169722922,"vramReservedGb":31.72990976,"vramReservedPct":11.039230354782568,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":130582.99918627474,"meanTps":130578.2632952053,"stepMs":250.93618774414062,"jitter":0.0005531795545416656,"achievedTflops":267.81350458366967,"nominalPeakTflops":2250.0,"mfuNominalPct":11.902822425940874,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.902822425940874,"vramAllocatedGb":57.621665792,"vramAllocatedPct":20.047294395588686,"vramReservedGb":60.049850368,"vramReservedPct":20.89209001842988,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":141578.3510779551,"meanTps":141448.12052630892,"stepMs":462.8956298828125,"jitter":0.00034596558291431973,"achievedTflops":290.3639417967178,"nominalPeakTflops":2250.0,"mfuNominalPct":12.905064079854125,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.905064079854125,"vramAllocatedGb":111.769860096,"vramAllocatedPct":38.88612484732021,"vramReservedGb":116.505182208,"vramReservedPct":40.53360232185002,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":64,"tokensPerStep":131072,"status":"complete","stable":true,"tps":148585.1248586398,"meanTps":148556.71661574664,"stepMs":882.1340637207031,"jitter":0.00013284317051035727,"achievedTflops":304.7341787626595,"nominalPeakTflops":2250.0,"mfuNominalPct":13.543741278340422,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.543741278340422,"vramAllocatedGb":220.066248192,"vramAllocatedPct":76.5637855726521,"vramReservedGb":229.55425792,"vramReservedPct":79.86478219659617,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":29880.348129090544,"meanTps":29848.72077763661,"stepMs":68.54003143310547,"jitter":0.00251303408102627,"achievedTflops":61.28179625600868,"nominalPeakTflops":165.2,"mfuNominalPct":37.095518314775234,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":29.8184070664222,"vramAllocatedGb":6.857717248,"vramAllocatedPct":27.147398239863225,"vramReservedGb":6.887047168,"vramReservedPct":27.26350553764013,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":36786.59286611829,"meanTps":36778.08127903793,"stepMs":111.34491348266602,"jitter":0.0012279993549492484,"achievedTflops":75.44585756614538,"nominalPeakTflops":165.2,"mfuNominalPct":45.6694053063834,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":36.71033536589671,"vramAllocatedGb":10.241995776,"vramAllocatedPct":40.54461973963103,"vramReservedGb":10.578034688,"vramReservedPct":41.87488487571766,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":38079.314878488025,"meanTps":38082.5877042821,"stepMs":215.12992095947266,"jitter":0.0003830914142096736,"achievedTflops":78.09710937336826,"nominalPeakTflops":165.2,"mfuNominalPct":47.27427928169992,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":38.00037760442864,"vramAllocatedGb":17.010520064,"vramAllocatedPct":67.33893302166541,"vramReservedGb":17.613979648,"vramReservedPct":69.72782673892796,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.317260288,"vramAllocatedPct":96.26386234184194,"vramReservedGb":24.387780608,"vramReservedPct":96.54302859233162,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 353.69 MiB is free. Including non-PyTorch memory, this process has 23.17 GiB memory in use. Of the allocated memory 22.65 GiB is allocated by PyTorch, and 67.25 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":29876.038163712197,"meanTps":29872.021803517564,"stepMs":68.54991912841797,"jitter":0.0018091565530232415,"achievedTflops":61.27295692056837,"nominalPeakTflops":165.2,"mfuNominalPct":37.0901676274627,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":29.814106035472406,"vramAllocatedGb":6.857717248,"vramAllocatedPct":27.147398239863225,"vramReservedGb":6.887047168,"vramReservedPct":27.26350553764013,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":36777.89874473293,"meanTps":36781.647549050685,"stepMs":111.37123489379883,"jitter":0.0009371844804751284,"achievedTflops":75.42802673723197,"nominalPeakTflops":165.2,"mfuNominalPct":45.65861182641161,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":36.7016592671634,"vramAllocatedGb":10.241995776,"vramAllocatedPct":40.54461973963103,"vramReservedGb":10.578034688,"vramReservedPct":41.87488487571766,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":38084.799844275876,"meanTps":38088.431379957234,"stepMs":215.09893798828125,"jitter":0.0005722483181198696,"achievedTflops":78.1083585246308,"nominalPeakTflops":165.2,"mfuNominalPct":47.2810886952971,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":38.00585120004743,"vramAllocatedGb":17.010520064,"vramAllocatedPct":67.33893302166541,"vramReservedGb":17.613979648,"vramReservedPct":69.72782673892796,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.317260288,"vramAllocatedPct":96.26386234184194,"vramReservedGb":24.387780608,"vramReservedPct":96.54302859233162,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 353.69 MiB is free. Including non-PyTorch memory, this process has 23.17 GiB memory in use. Of the allocated memory 22.65 GiB is allocated by PyTorch, and 67.25 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":37252.41394011036,"meanTps":37242.75703762705,"stepMs":54.97630310058594,"jitter":0.00047075458370374976,"achievedTflops":76.40121297314431,"nominalPeakTflops":209.5,"mfuNominalPct":36.46835941438869,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":28.40755375742187,"vramAllocatedGb":6.857717248,"vramAllocatedPct":20.368131515865816,"vramReservedGb":6.887047168,"vramReservedPct":20.455244420394514,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":47899.08118714076,"meanTps":47900.138262599256,"stepMs":85.51312255859375,"jitter":0.0004588657090019107,"achievedTflops":98.23653062805599,"nominalPeakTflops":209.5,"mfuNominalPct":46.89094540718663,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":36.52637721524214,"vramAllocatedGb":10.241995776,"vramAllocatedPct":30.41978976478649,"vramReservedGb":10.578034688,"vramReservedPct":31.417860187719224,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":54840.65337562561,"meanTps":54750.13738756084,"stepMs":149.37823486328125,"jitter":0.0006453855458804868,"achievedTflops":112.47304523335127,"nominalPeakTflops":209.5,"mfuNominalPct":53.68641777248271,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":41.81980827778831,"vramAllocatedGb":17.010520064,"vramAllocatedPct":50.52300893826911,"vramReservedGb":17.613979648,"vramReservedPct":52.31534649418195,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":54242.5374960868,"meanTps":54229.04652979034,"stepMs":302.05076599121094,"jitter":0.0007759323864337004,"achievedTflops":111.2463655672032,"nominalPeakTflops":209.5,"mfuNominalPct":53.100890485538514,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":41.36370336527224,"vramAllocatedGb":30.54756864,"vramAllocatedPct":90.72944728523434,"vramReservedGb":31.72990976,"vramReservedPct":94.24112304524026,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.879334912,"vramAllocatedPct":97.65503496620899,"vramReservedGb":33.017561088,"vramReservedPct":98.06558104588649,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 11.88 MiB is free. Including non-PyTorch memory, this process has 31.34 GiB memory in use. Of the allocated memory 30.62 GiB is allocated by PyTorch, and 131.82 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":37161.10154961329,"meanTps":37136.72805997625,"stepMs":55.11139106750488,"jitter":0.0009650109238031205,"achievedTflops":76.21393980999656,"nominalPeakTflops":209.5,"mfuNominalPct":36.378968883053254,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":28.337921715698762,"vramAllocatedGb":6.857717248,"vramAllocatedPct":20.368131515865816,"vramReservedGb":6.887047168,"vramReservedPct":20.455244420394514,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":47843.4918854592,"meanTps":47791.03839035833,"stepMs":85.61248016357422,"jitter":0.0008600518259799869,"achievedTflops":98.12252217524461,"nominalPeakTflops":209.5,"mfuNominalPct":46.836526097968786,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":36.48398651061841,"vramAllocatedGb":10.241995776,"vramAllocatedPct":30.41978976478649,"vramReservedGb":10.578034688,"vramReservedPct":31.417860187719224,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":54841.32561188829,"meanTps":54835.065209385764,"stepMs":149.37640380859375,"jitter":0.00033053413585041653,"achievedTflops":112.4744239269832,"nominalPeakTflops":209.5,"mfuNominalPct":53.68707586013518,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":41.820320904642514,"vramAllocatedGb":17.010520064,"vramAllocatedPct":50.52300893826911,"vramReservedGb":17.613979648,"vramReservedPct":52.31534649418195,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":54248.084200740916,"meanTps":54231.3157401245,"stepMs":302.01988220214844,"jitter":0.0003063598437340144,"achievedTflops":111.25774133909974,"nominalPeakTflops":209.5,"mfuNominalPct":53.106320448257634,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":41.36793311293078,"vramAllocatedGb":30.54756864,"vramAllocatedPct":90.72944728523434,"vramReservedGb":31.72990976,"vramReservedPct":94.24112304524026,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.879334912,"vramAllocatedPct":97.65503496620899,"vramReservedGb":33.017561088,"vramReservedPct":98.06558104588649,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 11.88 MiB is free. Including non-PyTorch memory, this process has 31.34 GiB memory in use. Of the allocated memory 30.62 GiB is allocated by PyTorch, and 131.82 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":35357.1529028139,"meanTps":34923.13810054945,"stepMs":57.92321586608887,"jitter":0.02183151044898674,"achievedTflops":72.51421004273071,"nominalPeakTflops":989.5,"mfuNominalPct":7.328368877486681,"configuredPeakTflops":989.5,"mfuConfiguredPct":7.328368877486681,"vramAllocatedGb":6.907786752,"vramAllocatedPct":8.125135742416347,"vramReservedGb":6.949961728,"vramReservedPct":8.174743151740895,"warmupSteps":16,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":64140.54769007744,"meanTps":63524.15935946666,"stepMs":63.85976028442383,"jitter":0.005206450954463418,"achievedTflops":131.5462577045875,"nominalPeakTflops":989.5,"mfuNominalPct":13.294215028255431,"configuredPeakTflops":989.5,"mfuConfiguredPct":13.294215028255431,"vramAllocatedGb":10.29206528,"vramAllocatedPct":12.105820644448624,"vramReservedGb":10.619977728,"vramReservedPct":12.491520615695803,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":79379.5979018491,"meanTps":79265.7239875753,"stepMs":103.20032119750977,"jitter":0.0005583312821390478,"achievedTflops":162.80012282618176,"nominalPeakTflops":989.5,"mfuNominalPct":16.452766329073448,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.452766329073448,"vramAllocatedGb":17.060589568,"vramAllocatedPct":20.06715190585725,"vramReservedGb":17.676894208,"vramReservedPct":20.79206699638624,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":89172.21962950664,"meanTps":89145.40134144365,"stepMs":183.7343521118164,"jitter":0.0004775485147283108,"achievedTflops":182.88387308684977,"nominalPeakTflops":989.5,"mfuNominalPct":18.482453065876683,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.482453065876683,"vramAllocatedGb":30.597638144,"vramAllocatedPct":35.9898144286745,"vramReservedGb":31.72990976,"vramReservedPct":37.321624588364436,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":95658.73727278093,"meanTps":95596.00041807043,"stepMs":342.55104064941406,"jitter":0.0002347785299623487,"achievedTflops":196.1871134275854,"nominalPeakTflops":989.5,"mfuNominalPct":19.826893726890894,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.826893726890894,"vramAllocatedGb":57.671735296,"vramAllocatedPct":67.835139474309,"vramReservedGb":60.007907328,"vramReservedPct":70.58301163063186,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.1045888,"vramAllocatedPct":98.92621545710819,"vramReservedGb":84.188069888,"vramReservedPct":99.0244082931462,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 40.19 MiB is free. Including non-PyTorch memory, this process has 79.13 GiB memory in use. Of the allocated memory 78.33 GiB is allocated by PyTorch, and 79.61 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":34271.4879157574,"meanTps":34271.51318419272,"stepMs":59.75812911987305,"jitter":0.01093785583171909,"achievedTflops":70.28761280726191,"nominalPeakTflops":989.5,"mfuNominalPct":7.103346418116414,"configuredPeakTflops":989.5,"mfuConfiguredPct":7.103346418116414,"vramAllocatedGb":6.907786752,"vramAllocatedPct":8.125135742416347,"vramReservedGb":6.949961728,"vramReservedPct":8.174743151740895,"warmupSteps":15,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":63492.347805539925,"meanTps":63583.465257321455,"stepMs":64.51171112060547,"jitter":0.003064553751390042,"achievedTflops":130.21686043365267,"nominalPeakTflops":989.5,"mfuNominalPct":13.159864621895167,"configuredPeakTflops":989.5,"mfuConfiguredPct":13.159864621895167,"vramAllocatedGb":10.29206528,"vramAllocatedPct":12.105820644448624,"vramReservedGb":10.619977728,"vramReservedPct":12.491520615695803,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":79207.27238579617,"meanTps":79207.08015405925,"stepMs":103.42484664916992,"jitter":0.0005546584244740274,"achievedTflops":162.4466993279399,"nominalPeakTflops":989.5,"mfuNominalPct":16.417048946734706,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.417048946734706,"vramAllocatedGb":17.060589568,"vramAllocatedPct":20.06715190585725,"vramReservedGb":17.676894208,"vramReservedPct":20.79206699638624,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":88998.41306021767,"meanTps":88981.25880066569,"stepMs":184.09317016601562,"jitter":0.0005539137220642318,"achievedTflops":182.5274120870948,"nominalPeakTflops":989.5,"mfuNominalPct":18.446428710166227,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.446428710166227,"vramAllocatedGb":30.597638144,"vramAllocatedPct":35.9898144286745,"vramReservedGb":31.72990976,"vramReservedPct":37.321624588364436,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":95604.40106803505,"meanTps":95595.56721114652,"stepMs":342.7457275390625,"jitter":0.0002310809403039348,"achievedTflops":196.0756749592591,"nominalPeakTflops":989.5,"mfuNominalPct":19.81563162802012,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.81563162802012,"vramAllocatedGb":57.671735296,"vramAllocatedPct":67.835139474309,"vramReservedGb":60.007907328,"vramReservedPct":70.58301163063186,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.1045888,"vramAllocatedPct":98.92621545710819,"vramReservedGb":84.188069888,"vramReservedPct":99.0244082931462,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 40.19 MiB is free. Including non-PyTorch memory, this process has 79.13 GiB memory in use. Of the allocated memory 78.33 GiB is allocated by PyTorch, and 79.61 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":33466.8429968442,"meanTps":33376.90140873791,"stepMs":61.19489669799805,"jitter":0.0018246602594771497,"achievedTflops":68.63736141908397,"nominalPeakTflops":989.5,"mfuNominalPct":6.936570128255076,"configuredPeakTflops":989.5,"mfuConfiguredPct":6.936570128255076,"vramAllocatedGb":6.907786752,"vramAllocatedPct":4.601486686048263,"vramReservedGb":6.949961728,"vramReservedPct":4.6295807192771,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":65322.253418630804,"meanTps":65073.244987203114,"stepMs":62.704511642456055,"jitter":0.0024752374637884585,"achievedTflops":133.9698255083797,"nominalPeakTflops":989.5,"mfuNominalPct":13.53914355819906,"configuredPeakTflops":989.5,"mfuConfiguredPct":13.53914355819906,"vramAllocatedGb":10.29206528,"vramAllocatedPct":6.855857463195122,"vramReservedGb":10.619977728,"vramReservedPct":7.074289910204958,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":85646.39784859396,"meanTps":85581.49112706461,"stepMs":95.64908981323242,"jitter":0.0004712019282251797,"achievedTflops":175.65274274394287,"nominalPeakTflops":989.5,"mfuNominalPct":17.75166677553743,"configuredPeakTflops":989.5,"mfuConfiguredPct":17.75166677553743,"vramAllocatedGb":17.060589568,"vramAllocatedPct":11.364577189728205,"vramReservedGb":17.676894208,"vramReservedPct":11.775116440189098,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":98161.56268364009,"meanTps":98156.93504206593,"stepMs":166.90850830078125,"jitter":0.00047459124254149276,"achievedTflops":201.32017400070862,"nominalPeakTflops":989.5,"mfuNominalPct":20.345646690319214,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.345646690319214,"vramAllocatedGb":30.597638144,"vramAllocatedPct":20.382016642794373,"vramReservedGb":31.72990976,"vramReservedPct":21.136257176422003,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":106661.39098936583,"meanTps":106663.11222900973,"stepMs":307.21519470214844,"jitter":0.0007487501300682861,"achievedTflops":218.75252599982818,"nominalPeakTflops":989.5,"mfuNominalPct":22.107380090937664,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.107380090937664,"vramAllocatedGb":57.671735296,"vramAllocatedPct":38.41689554892671,"vramReservedGb":60.007907328,"vramReservedPct":39.97309073669129,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":111363.276203408,"meanTps":111314.30240804085,"stepMs":588.4884338378906,"jitter":0.0026734114277171123,"achievedTflops":228.3956523269123,"nominalPeakTflops":989.5,"mfuNominalPct":23.081925449915342,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.081925449915342,"vramAllocatedGb":111.8199296,"vramAllocatedPct":74.48665336119137,"vramReservedGb":116.379353088,"vramReservedPct":77.5238239093432,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.661396992,"vramAllocatedPct":99.02787441867217,"vramReservedGb":148.849557504,"vramReservedPct":99.15321385399237,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 463.44 MiB is free. Including non-PyTorch memory, this process has 139.35 GiB memory in use. Of the allocated memory 138.45 GiB is allocated by PyTorch, and 179.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":33000.39963442501,"meanTps":32912.17668743333,"stepMs":62.05985450744629,"jitter":0.014557096676647743,"achievedTflops":67.68072975678713,"nominalPeakTflops":989.5,"mfuNominalPct":6.8398918399987,"configuredPeakTflops":989.5,"mfuConfiguredPct":6.8398918399987,"vramAllocatedGb":6.907786752,"vramAllocatedPct":4.601486686048263,"vramReservedGb":6.949961728,"vramReservedPct":4.6295807192771,"warmupSteps":27,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":65337.540808891295,"meanTps":65149.904305969416,"stepMs":62.68984031677246,"jitter":0.002481906371366927,"achievedTflops":134.00117851441505,"nominalPeakTflops":989.5,"mfuNominalPct":13.542312128793839,"configuredPeakTflops":989.5,"mfuConfiguredPct":13.542312128793839,"vramAllocatedGb":10.29206528,"vramAllocatedPct":6.855857463195122,"vramReservedGb":10.619977728,"vramReservedPct":7.074289910204958,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":86298.96189294146,"meanTps":86300.69350881268,"stepMs":94.92582321166992,"jitter":0.00048730392555768737,"achievedTflops":176.99109049802297,"nominalPeakTflops":989.5,"mfuNominalPct":17.886921727945726,"configuredPeakTflops":989.5,"mfuConfiguredPct":17.886921727945726,"vramAllocatedGb":17.060589568,"vramAllocatedPct":11.364577189728205,"vramReservedGb":17.676894208,"vramReservedPct":11.775116440189098,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":98418.83599167803,"meanTps":98422.30966548952,"stepMs":166.47219848632812,"jitter":0.0002635676680364901,"achievedTflops":201.84781746648005,"nominalPeakTflops":989.5,"mfuNominalPct":20.398970941534113,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.398970941534113,"vramAllocatedGb":30.597638144,"vramAllocatedPct":20.382016642794373,"vramReservedGb":31.72990976,"vramReservedPct":21.136257176422003,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":106842.16185504281,"meanTps":106744.82452165453,"stepMs":306.6954040527344,"jitter":0.0011981275140438186,"achievedTflops":219.123270119394,"nominalPeakTflops":989.5,"mfuNominalPct":22.144847915047396,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.144847915047396,"vramAllocatedGb":57.671735296,"vramAllocatedPct":38.41689554892671,"vramReservedGb":60.007907328,"vramReservedPct":39.97309073669129,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":111300.32934264411,"meanTps":111272.03050125595,"stepMs":588.8212585449219,"jitter":0.0024525204831740864,"achievedTflops":228.26655420932607,"nominalPeakTflops":989.5,"mfuNominalPct":23.0688786467232,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.0688786467232,"vramAllocatedGb":111.8199296,"vramAllocatedPct":74.48665336119137,"vramReservedGb":116.379353088,"vramReservedPct":77.5238239093432,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.661396992,"vramAllocatedPct":99.02787441867217,"vramReservedGb":148.849557504,"vramReservedPct":99.15321385399237,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 463.44 MiB is free. Including non-PyTorch memory, this process has 139.35 GiB memory in use. Of the allocated memory 138.45 GiB is allocated by PyTorch, and 179.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":22937.764921185775,"meanTps":22850.412093694234,"stepMs":89.28507232666016,"jitter":0.013182195066075365,"achievedTflops":47.043208144546924,"nominalPeakTflops":364.2,"mfuNominalPct":12.916861105037597,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":10.420849297300862,"vramAllocatedGb":6.857717248,"vramAllocatedPct":13.479054478571308,"vramReservedGb":6.887047168,"vramReservedPct":13.536703339735341,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":35234.64678158461,"meanTps":35119.19379964351,"stepMs":116.24921417236328,"jitter":0.010483469736891413,"achievedTflops":72.26296145858258,"nominalPeakTflops":364.2,"mfuNominalPct":19.84155998313635,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.007442111998888,"vramAllocatedGb":10.241995776,"vramAllocatedPct":20.130958166037416,"vramReservedGb":10.578034688,"vramReservedPct":20.791452998058787,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":36277.89399368309,"meanTps":36298.96669856222,"stepMs":225.81244659423828,"jitter":0.002813753708946668,"achievedTflops":74.40256380927359,"nominalPeakTflops":364.2,"mfuNominalPct":20.42904003549522,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.481399448926094,"vramAllocatedGb":17.010520064,"vramAllocatedPct":33.434701134456326,"vramReservedGb":17.613979648,"vramReservedPct":34.62081953423786,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":34683.38260699182,"meanTps":34706.19997828822,"stepMs":472.3876037597656,"jitter":0.002616761922469234,"achievedTflops":71.13237025245996,"nominalPeakTflops":364.2,"mfuNominalPct":19.53112857014277,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.7569974454775,"vramAllocatedGb":30.54756864,"vramAllocatedPct":60.042187071294144,"vramReservedGb":31.72990976,"vramReservedPct":62.36611496047372,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":49.237152256,"vramAllocatedPct":96.77713933479664,"vramReservedGb":49.564090368,"vramReservedPct":97.4197462641002,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.38 GiB of which 866.50 MiB is free. Process 2308332 has 46.53 GiB memory in use. Of the allocated memory 45.86 GiB is allocated by PyTorch, and 171.79 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":24571.699127797016,"meanTps":24519.76773493423,"stepMs":83.34791946411133,"jitter":0.011553733482715146,"achievedTflops":50.39425421377894,"nominalPeakTflops":364.2,"mfuNominalPct":13.83697260125726,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":11.163161470579528,"vramAllocatedGb":6.857717248,"vramAllocatedPct":13.479054478571308,"vramReservedGb":6.887047168,"vramReservedPct":13.536703339735341,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":35247.82219323517,"meanTps":35220.01640263185,"stepMs":116.20576095581055,"jitter":0.004213557405376808,"achievedTflops":72.28998299423773,"nominalPeakTflops":364.2,"mfuNominalPct":19.848979405337104,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.013427829426554,"vramAllocatedGb":10.241995776,"vramAllocatedPct":20.130958166037416,"vramReservedGb":10.578034688,"vramReservedPct":20.791452998058787,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":36040.46861055191,"meanTps":36049.299110508706,"stepMs":227.30004119873047,"jitter":0.002843311148570734,"achievedTflops":73.91562657908501,"nominalPeakTflops":364.2,"mfuNominalPct":20.29533953297227,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.3735347923013,"vramAllocatedGb":17.010520064,"vramAllocatedPct":33.434701134456326,"vramReservedGb":17.613979648,"vramReservedPct":34.62081953423786,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":34495.96818976856,"meanTps":34502.893467614726,"stepMs":474.9540557861328,"jitter":0.0015161520139582271,"achievedTflops":70.74800083072178,"nominalPeakTflops":364.2,"mfuNominalPct":19.425590563075723,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.671853256201185,"vramAllocatedGb":30.54756864,"vramAllocatedPct":60.042187071294144,"vramReservedGb":31.72990976,"vramReservedPct":62.36611496047372,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":49.237152256,"vramAllocatedPct":96.77713933479664,"vramReservedGb":49.564090368,"vramReservedPct":97.4197462641002,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.38 GiB of which 866.50 MiB is free. Process 2312978 has 46.53 GiB memory in use. Of the allocated memory 45.86 GiB is allocated by PyTorch, and 171.79 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15791.315794071497,"meanTps":15795.983057762296,"stepMs":129.69153594970703,"jitter":0.002874942521378847,"achievedTflops":32.386510121160214,"nominalPeakTflops":154.8,"mfuNominalPct":20.921518166124166,"configuredPeakTflops":180.6,"mfuConfiguredPct":17.93272985667786,"vramAllocatedGb":6.857717248,"vramAllocatedPct":13.437564007609963,"vramReservedGb":6.887047168,"vramReservedPct":13.495035417276645,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":19522.484910254247,"meanTps":19492.82930005122,"stepMs":209.80935668945312,"jitter":0.0016329547581807968,"achievedTflops":40.03878862162434,"nominalPeakTflops":154.8,"mfuNominalPct":25.864850530765075,"configuredPeakTflops":180.6,"mfuConfiguredPct":22.169871883512926,"vramAllocatedGb":10.241995776,"vramAllocatedPct":20.06899217750759,"vramReservedGb":10.578034688,"vramReservedPct":20.727453911310413,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":22213.03516962075,"meanTps":22208.964736164573,"stepMs":368.79246520996094,"jitter":0.0018318120724229695,"achievedTflops":45.556855281983154,"nominalPeakTflops":154.8,"mfuNominalPct":29.429493076216506,"configuredPeakTflops":180.6,"mfuConfiguredPct":25.22527977961415,"vramAllocatedGb":17.010520064,"vramAllocatedPct":33.331784309042064,"vramReservedGb":17.613979648,"vramReservedPct":34.514251665562284,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23961.228996916292,"meanTps":23954.58332584108,"stepMs":683.7712707519531,"jitter":0.0008183944731852447,"achievedTflops":49.142237134881874,"nominalPeakTflops":154.8,"mfuNominalPct":31.745631224083894,"configuredPeakTflops":180.6,"mfuConfiguredPct":27.21054104921477,"vramAllocatedGb":30.54756864,"vramAllocatedPct":59.857368572111014,"vramReservedGb":31.72990976,"vramReservedPct":62.17414307655166,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.237152256,"vramAllocatedPct":96.47924536191628,"vramReservedGb":49.564090368,"vramReservedPct":97.1198742545421,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.18 GiB is free. Process 1348507 has 46.35 GiB memory in use. Of the allocated memory 45.86 GiB is allocated by PyTorch, and 171.79 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15700.214638738638,"meanTps":15669.569304463594,"stepMs":130.44407653808594,"jitter":0.004638944057885198,"achievedTflops":32.19967018155589,"nominalPeakTflops":154.8,"mfuNominalPct":20.800820530720856,"configuredPeakTflops":180.6,"mfuConfiguredPct":17.82927474061788,"vramAllocatedGb":6.857717248,"vramAllocatedPct":13.437564007609963,"vramReservedGb":6.887047168,"vramReservedPct":13.495035417276645,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":19397.394184354096,"meanTps":19370.885285947148,"stepMs":211.16238403320312,"jitter":0.002417961861840292,"achievedTflops":39.782239255297974,"nominalPeakTflops":154.8,"mfuNominalPct":25.699120965954762,"configuredPeakTflops":180.6,"mfuConfiguredPct":22.02781797081837,"vramAllocatedGb":10.241995776,"vramAllocatedPct":20.06899217750759,"vramReservedGb":10.578034688,"vramReservedPct":20.727453911310413,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":22116.471591894264,"meanTps":22093.147137723186,"stepMs":370.4026641845703,"jitter":0.0014650889862366629,"achievedTflops":45.35881242550705,"nominalPeakTflops":154.8,"mfuNominalPct":29.301558414410238,"configuredPeakTflops":180.6,"mfuConfiguredPct":25.11562149806592,"vramAllocatedGb":17.010520064,"vramAllocatedPct":33.331784309042064,"vramReservedGb":17.613979648,"vramReservedPct":34.514251665562284,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23834.03298082595,"meanTps":23835.536524708845,"stepMs":687.4203796386719,"jitter":0.0008109566044824705,"achievedTflops":48.881370015497964,"nominalPeakTflops":154.8,"mfuNominalPct":31.577112413112374,"configuredPeakTflops":180.6,"mfuConfiguredPct":27.066096354096324,"vramAllocatedGb":30.54756864,"vramAllocatedPct":59.857368572111014,"vramReservedGb":31.72990976,"vramReservedPct":62.17414307655166,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.237152256,"vramAllocatedPct":96.47924536191628,"vramReservedGb":49.564090368,"vramReservedPct":97.1198742545421,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.18 GiB is free. Process 1380474 has 46.35 GiB memory in use. Of the allocated memory 45.86 GiB is allocated by PyTorch, and 171.79 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":38871.09952018489,"meanTps":38786.434370994815,"stepMs":52.68695831298828,"jitter":0.0012813210156412342,"achievedTflops":79.72098553710893,"nominalPeakTflops":468.0,"mfuNominalPct":17.03439861904037,"configuredPeakTflops":468.0,"mfuConfiguredPct":17.03439861904037,"vramAllocatedGb":6.857717248,"vramAllocatedPct":6.72496967784041,"vramReservedGb":6.887047168,"vramReservedPct":6.753731846871367,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":52086.35257803092,"meanTps":52088.509683052376,"stepMs":78.63864135742188,"jitter":0.002497700440967995,"achievedTflops":106.82423218817682,"nominalPeakTflops":468.0,"mfuNominalPct":22.82569063849932,"configuredPeakTflops":468.0,"mfuConfiguredPct":22.82569063849932,"vramAllocatedGb":10.241995776,"vramAllocatedPct":10.043737375473892,"vramReservedGb":10.578034688,"vramReservedPct":10.373271448117897,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":66068.43683841648,"meanTps":66057.81643131592,"stepMs":123.99264144897461,"jitter":0.0006049189441017388,"achievedTflops":135.50017783571414,"nominalPeakTflops":468.0,"mfuNominalPct":28.953029452075672,"configuredPeakTflops":468.0,"mfuConfiguredPct":28.953029452075672,"vramAllocatedGb":17.010520064,"vramAllocatedPct":16.681240637044112,"vramReservedGb":17.613979648,"vramReservedPct":17.273018812994096,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":68173.33091050295,"meanTps":68172.22371925368,"stepMs":240.32858276367188,"jitter":0.000693660870032293,"achievedTflops":139.81711849212175,"nominalPeakTflops":468.0,"mfuNominalPct":29.875452669256784,"configuredPeakTflops":468.0,"mfuConfiguredPct":29.875452669256784,"vramAllocatedGb":30.54756864,"vramAllocatedPct":29.95624716018455,"vramReservedGb":31.72990976,"vramReservedPct":31.11570123117046,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":61812.45954095459,"meanTps":61808.50714651087,"stepMs":530.1196594238281,"jitter":0.0002547677701884196,"achievedTflops":126.77156689428612,"nominalPeakTflops":468.0,"mfuNominalPct":27.08794164407823,"configuredPeakTflops":468.0,"mfuConfiguredPct":27.08794164407823,"vramAllocatedGb":57.621665792,"vramAllocatedPct":56.50626020646543,"vramReservedGb":60.049850368,"vramReservedPct":58.887441444371106,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":99.19309056,"vramAllocatedPct":97.27297031119558,"vramReservedGb":99.748937728,"vramReservedPct":97.81805772368749,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 94.97 GiB of which 1.79 GiB is free. Including non-PyTorch memory, this process has 93.17 GiB memory in use. Of the allocated memory 92.38 GiB is allocated by PyTorch, and 150.10 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":38885.907426172984,"meanTps":38885.85109562602,"stepMs":52.66689491271973,"jitter":0.0008179567017560556,"achievedTflops":79.75135516579665,"nominalPeakTflops":468.0,"mfuNominalPct":17.040887855939456,"configuredPeakTflops":468.0,"mfuConfiguredPct":17.040887855939456,"vramAllocatedGb":6.857717248,"vramAllocatedPct":6.72496967784041,"vramReservedGb":6.887047168,"vramReservedPct":6.753731846871367,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":52251.402915229555,"meanTps":52245.328335571634,"stepMs":78.39023971557617,"jitter":0.0005849008524817427,"achievedTflops":107.16273497577805,"nominalPeakTflops":468.0,"mfuNominalPct":22.89802029396967,"configuredPeakTflops":468.0,"mfuConfiguredPct":22.89802029396967,"vramAllocatedGb":10.241995776,"vramAllocatedPct":10.043737375473892,"vramReservedGb":10.578034688,"vramReservedPct":10.373271448117897,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":66127.85520158894,"meanTps":66095.82282483824,"stepMs":123.88122940063477,"jitter":0.0003006314049626477,"achievedTflops":135.6220393351207,"nominalPeakTflops":468.0,"mfuNominalPct":28.979068234000145,"configuredPeakTflops":468.0,"mfuConfiguredPct":28.979068234000145,"vramAllocatedGb":17.010520064,"vramAllocatedPct":16.681240637044112,"vramReservedGb":17.613979648,"vramReservedPct":17.273018812994096,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":68222.3053215233,"meanTps":68198.93813809981,"stepMs":240.15605926513672,"jitter":0.00023898518746663314,"achievedTflops":139.9175604235524,"nominalPeakTflops":468.0,"mfuNominalPct":29.89691462041718,"configuredPeakTflops":468.0,"mfuConfiguredPct":29.89691462041718,"vramAllocatedGb":30.54756864,"vramAllocatedPct":29.95624716018455,"vramReservedGb":31.72990976,"vramReservedPct":31.11570123117046,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":61801.611947961355,"meanTps":61798.881615612736,"stepMs":530.2127075195312,"jitter":0.00025113901741739527,"achievedTflops":126.74931949673883,"nominalPeakTflops":468.0,"mfuNominalPct":27.083187926653597,"configuredPeakTflops":468.0,"mfuConfiguredPct":27.083187926653597,"vramAllocatedGb":57.621665792,"vramAllocatedPct":56.50626020646543,"vramReservedGb":60.049850368,"vramReservedPct":58.887441444371106,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":99.19309056,"vramAllocatedPct":97.27297031119558,"vramReservedGb":99.748937728,"vramReservedPct":97.81805772368749,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 94.97 GiB of which 1.79 GiB is free. Including non-PyTorch memory, this process has 93.17 GiB memory in use. Of the allocated memory 92.38 GiB is allocated by PyTorch, and 150.10 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":39607.75442186366,"meanTps":39588.63870604135,"stepMs":413.6563720703125,"jitter":0.0024171687117641627,"achievedTflops":103.90812618854076,"nominalPeakTflops":312.0,"mfuNominalPct":33.30388659889127,"configuredPeakTflops":312.0,"mfuConfiguredPct":33.30388659889127,"vramAllocatedGb":30.555664384,"vramAllocatedPct":35.95873817295026,"vramReservedGb":31.736201216,"vramReservedPct":37.34802607426786,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":41701.5732302873,"meanTps":41694.25866482508,"stepMs":785.7737121582031,"jitter":0.0009805408938158978,"achievedTflops":109.40111088654542,"nominalPeakTflops":312.0,"mfuNominalPct":35.06445861748251,"configuredPeakTflops":312.0,"mfuConfiguredPct":35.06445861748251,"vramAllocatedGb":57.631041536,"vramAllocatedPct":67.8217795294477,"vramReservedGb":60.045656064,"vramReservedPct":70.66336367927293,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.237512192,"vramAllocatedPct":99.13299894859958,"vramReservedGb":84.393590784,"vramReservedPct":99.3166764871927,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 68.75 MiB is free. Process 764857 has 79.06 GiB memory in use. Of the allocated memory 78.43 GiB is allocated by PyTorch, and 132.88 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":39627.69296905079,"meanTps":39590.37267041698,"stepMs":413.4482421875,"jitter":0.002352670301284216,"achievedTflops":103.96043354873768,"nominalPeakTflops":312.0,"mfuNominalPct":33.320651778441565,"configuredPeakTflops":312.0,"mfuConfiguredPct":33.320651778441565,"vramAllocatedGb":30.555664384,"vramAllocatedPct":35.95873817295026,"vramReservedGb":31.736201216,"vramReservedPct":37.34802607426786,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":41642.09291979748,"meanTps":41651.33923855745,"stepMs":786.8960876464844,"jitter":0.0012222460638081558,"achievedTflops":109.24506852316674,"nominalPeakTflops":312.0,"mfuNominalPct":35.01444503947652,"configuredPeakTflops":312.0,"mfuConfiguredPct":35.01444503947652,"vramAllocatedGb":57.631041536,"vramAllocatedPct":67.8217795294477,"vramReservedGb":60.045656064,"vramReservedPct":70.66336367927293,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.237512192,"vramAllocatedPct":99.13299894859958,"vramReservedGb":84.393590784,"vramReservedPct":99.3166764871927,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 68.75 MiB is free. Process 852916 has 79.06 GiB memory in use. Of the allocated memory 78.43 GiB is allocated by PyTorch, and 132.88 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":114360.63361226003,"meanTps":114191.05797818479,"stepMs":286.53216552734375,"jitter":0.0013131524836389823,"achievedTflops":300.01698712373155,"nominalPeakTflops":2250.0,"mfuNominalPct":13.334088316610291,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.334088316610291,"vramAllocatedGb":57.631041536,"vramAllocatedPct":30.09406599662435,"vramReservedGb":60.056141824,"vramReservedPct":31.360417014589487,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":122088.41906772403,"meanTps":122046.94639312325,"stepMs":536.7912902832031,"jitter":0.0017020387585215948,"achievedTflops":320.2902825423952,"nominalPeakTflops":2250.0,"mfuNominalPct":14.235123668550896,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.235123668550896,"vramAllocatedGb":111.78179584,"vramAllocatedPct":58.37077817045526,"vramReservedGb":116.471627776,"vramReservedPct":60.81973810651503,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.537062912,"vramAllocatedPct":97.40685804860128,"vramReservedGb":187.305033728,"vramReservedPct":97.80788089677849,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.68 GiB is free. Including non-PyTorch memory, this process has 174.65 GiB memory in use. Of the allocated memory 173.73 GiB is allocated by PyTorch, and 112.39 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":114342.79245168895,"meanTps":114162.58905095296,"stepMs":286.5768737792969,"jitter":0.0014158078157501086,"achievedTflops":299.9701821081221,"nominalPeakTflops":2250.0,"mfuNominalPct":13.332008093694316,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.332008093694316,"vramAllocatedGb":57.631041536,"vramAllocatedPct":30.09406599662435,"vramReservedGb":60.056141824,"vramReservedPct":31.360417014589487,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":122311.41503680362,"meanTps":122290.53702354981,"stepMs":535.8126220703125,"jitter":0.00033312569186925615,"achievedTflops":320.8752966042342,"nominalPeakTflops":2250.0,"mfuNominalPct":14.26112429352152,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.26112429352152,"vramAllocatedGb":111.78179584,"vramAllocatedPct":58.37077817045526,"vramReservedGb":116.471627776,"vramReservedPct":60.81973810651503,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.537062912,"vramAllocatedPct":97.40685804860128,"vramReservedGb":187.305033728,"vramReservedPct":97.80788089677849,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.68 GiB is free. Including non-PyTorch memory, this process has 174.65 GiB memory in use. Of the allocated memory 173.73 GiB is allocated by PyTorch, and 112.39 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":123060.60019991269,"meanTps":122990.73125628174,"stepMs":532.5506286621094,"jitter":0.00030415585831081404,"achievedTflops":322.8407305856151,"nominalPeakTflops":2250.0,"mfuNominalPct":14.348476914916228,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.348476914916228,"vramAllocatedGb":111.78179584,"vramAllocatedPct":38.89027744114945,"vramReservedGb":116.425490432,"vramReservedPct":40.50587656154059,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":32,"tokensPerStep":131072,"status":"complete","stable":true,"tps":128594.74976723274,"meanTps":128590.8981130359,"stepMs":1019.2640075683594,"jitter":0.00014528921508789885,"achievedTflops":337.3591782982156,"nominalPeakTflops":2250.0,"mfuNominalPct":14.993741257698472,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.993741257698472,"vramAllocatedGb":220.083304448,"vramAllocatedPct":76.56971965631001,"vramReservedGb":229.266948096,"vramReservedPct":79.76482353442793,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.056800768,"vramAllocatedPct":99.52271979704788,"vramReservedGb":286.192041984,"vramReservedPct":99.56977189861948,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 325.62 MiB is free. Including non-PyTorch memory, this process has 267.36 GiB memory in use. Of the allocated memory 266.41 GiB is allocated by PyTorch, and 128.98 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":123152.33661474445,"meanTps":123142.2444931358,"stepMs":532.1539306640625,"jitter":0.0004718780429898117,"achievedTflops":323.08139454416465,"nominalPeakTflops":2250.0,"mfuNominalPct":14.359173090851762,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.359173090851762,"vramAllocatedGb":111.78179584,"vramAllocatedPct":38.89027744114945,"vramReservedGb":116.425490432,"vramReservedPct":40.50587656154059,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":32,"tokensPerStep":131072,"status":"complete","stable":true,"tps":128661.58596817743,"meanTps":128649.71833854205,"stepMs":1018.7345275878906,"jitter":0.00014483176852792622,"achievedTflops":337.53451831693405,"nominalPeakTflops":2250.0,"mfuNominalPct":15.00153414741929,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.00153414741929,"vramAllocatedGb":220.083304448,"vramAllocatedPct":76.56971965631001,"vramReservedGb":229.266948096,"vramReservedPct":79.76482353442793,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.056800768,"vramAllocatedPct":99.52271979704788,"vramReservedGb":286.192041984,"vramReservedPct":99.56977189861948,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 325.62 MiB is free. Including non-PyTorch memory, this process has 267.36 GiB memory in use. Of the allocated memory 266.41 GiB is allocated by PyTorch, and 128.98 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":33495.36710702765,"meanTps":33493.67990737101,"stepMs":122.28556823730469,"jitter":0.0008502081329835003,"achievedTflops":87.8727128788525,"nominalPeakTflops":165.2,"mfuNominalPct":53.19171481770732,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":42.75697650418579,"vramAllocatedGb":10.249098752,"vramAllocatedPct":40.572738034857686,"vramReservedGb":10.311696384,"vramReservedPct":40.82054102575411,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":34789.88834037539,"meanTps":34788.868129311864,"stepMs":235.47071838378906,"jitter":0.0003102719000734133,"achievedTflops":91.26879724747786,"nominalPeakTflops":165.2,"mfuNominalPct":55.24745596094302,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":44.40943828439439,"vramAllocatedGb":17.017975808,"vramAllocatedPct":67.3684478068662,"vramReservedGb":17.620271104,"vramReservedPct":69.75273249916332,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.324076032,"vramAllocatedPct":96.29084358209691,"vramReservedGb":24.354226176,"vramReservedPct":96.41019787107636,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 385.69 MiB is free. Including non-PyTorch memory, this process has 23.14 GiB memory in use. Of the allocated memory 22.65 GiB is allocated by PyTorch, and 28.75 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":33456.459984507244,"meanTps":33437.319876963425,"stepMs":122.42777633666992,"jitter":0.0016998731932088742,"achievedTflops":87.7706428106173,"nominalPeakTflops":165.2,"mfuNominalPct":53.129929062117014,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":42.70731139921368,"vramAllocatedGb":10.249098752,"vramAllocatedPct":40.572738034857686,"vramReservedGb":10.311696384,"vramReservedPct":40.82054102575411,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":34736.578557318004,"meanTps":34720.910837881376,"stepMs":235.83209228515625,"jitter":0.0012987565455696913,"achievedTflops":91.12894282387153,"nominalPeakTflops":165.2,"mfuNominalPct":55.16279831953483,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":44.34138812287967,"vramAllocatedGb":17.017975808,"vramAllocatedPct":67.3684478068662,"vramReservedGb":17.620271104,"vramReservedPct":69.75273249916332,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.324076032,"vramAllocatedPct":96.29084358209691,"vramReservedGb":24.354226176,"vramReservedPct":96.41019787107636,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 385.69 MiB is free. Including non-PyTorch memory, this process has 23.14 GiB memory in use. Of the allocated memory 22.65 GiB is allocated by PyTorch, and 28.75 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":43576.7157536658,"meanTps":43577.00844511606,"stepMs":93.99515151977539,"jitter":0.00034320858156355053,"achievedTflops":114.32041390649037,"nominalPeakTflops":209.5,"mfuNominalPct":54.568216661809245,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":42.50669822167471,"vramAllocatedGb":10.249098752,"vramAllocatedPct":30.44088634023428,"vramReservedGb":10.332667904,"vramReservedPct":30.689095389550477,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":49825.451772377324,"meanTps":49818.37596675875,"stepMs":164.4139633178711,"jitter":0.00048158550968774246,"achievedTflops":130.71352834149462,"nominalPeakTflops":209.5,"mfuNominalPct":62.3930922871096,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":48.60199777834072,"vramAllocatedGb":17.017975808,"vramAllocatedPct":50.545153271266344,"vramReservedGb":17.620271104,"vramReservedPct":52.33403277105807,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":49794.87711968378,"meanTps":49793.54749360722,"stepMs":329.0298309326172,"jitter":0.00033594041431229976,"achievedTflops":130.6333179151116,"nominalPeakTflops":209.5,"mfuNominalPct":62.35480568740411,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":48.57217388012352,"vramAllocatedGb":30.555664384,"vramAllocatedPct":90.75349248461302,"vramReservedGb":31.715229696,"vramReservedPct":94.19752173252931,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":43507.6551567732,"meanTps":43505.42747457929,"stepMs":94.14435195922852,"jitter":0.00038651901139967996,"achievedTflops":114.13923834323727,"nominalPeakTflops":209.5,"mfuNominalPct":54.48173667934953,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":42.43933339391402,"vramAllocatedGb":10.249098752,"vramAllocatedPct":30.44088634023428,"vramReservedGb":10.332667904,"vramReservedPct":30.689095389550477,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":49791.988425448246,"meanTps":49785.0089132884,"stepMs":164.5244598388672,"jitter":0.0005728393683151283,"achievedTflops":130.62573963127485,"nominalPeakTflops":209.5,"mfuNominalPct":62.35118836815028,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":48.569356117196676,"vramAllocatedGb":17.017975808,"vramAllocatedPct":50.545153271266344,"vramReservedGb":17.620271104,"vramReservedPct":52.33403277105807,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":49780.61470846048,"meanTps":49778.2383572844,"stepMs":329.1240997314453,"jitter":0.0005052097461612073,"achievedTflops":130.59590149383817,"nominalPeakTflops":209.5,"mfuNominalPct":62.336945820447816,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":48.558261679552714,"vramAllocatedGb":30.555664384,"vramAllocatedPct":90.75349248461302,"vramReservedGb":31.715229696,"vramReservedPct":94.19752173252931,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":80282.83279768046,"meanTps":80276.76184171344,"stepMs":204.07849884033203,"jitter":0.0002465367105441079,"achievedTflops":210.61630084512078,"nominalPeakTflops":989.5,"mfuNominalPct":21.28512388530781,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.28512388530781,"vramAllocatedGb":30.605733888,"vramAllocatedPct":35.999336873604754,"vramReservedGb":31.757172736,"vramReservedPct":37.35369207809667,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":85964.82016563095,"meanTps":85943.06102987897,"stepMs":381.17918395996094,"jitter":0.0005476150935598589,"achievedTflops":225.5225905110855,"nominalPeakTflops":989.5,"mfuNominalPct":22.791570541797423,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.791570541797423,"vramAllocatedGb":57.68111104,"vramAllocatedPct":67.84616749173645,"vramReservedGb":60.108570624,"vramReservedPct":70.70141466964319,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.111420928,"vramAllocatedPct":98.93425160086922,"vramReservedGb":84.204847104,"vramReservedPct":99.0441421329814,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 24.19 MiB is free. Including non-PyTorch memory, this process has 79.15 GiB memory in use. Of the allocated memory 78.33 GiB is allocated by PyTorch, and 89.10 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":80238.3895287189,"meanTps":80184.83029762394,"stepMs":204.19153594970703,"jitter":0.0011075823370775585,"achievedTflops":210.4997070905165,"nominalPeakTflops":989.5,"mfuNominalPct":21.273340787318492,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.273340787318492,"vramAllocatedGb":30.605733888,"vramAllocatedPct":35.999336873604754,"vramReservedGb":31.757172736,"vramReservedPct":37.35369207809667,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":85864.6428659393,"meanTps":85841.84527499361,"stepMs":381.6239013671875,"jitter":0.0008648365236705932,"achievedTflops":225.25978249155685,"nominalPeakTflops":989.5,"mfuNominalPct":22.765010863219487,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.765010863219487,"vramAllocatedGb":57.68111104,"vramAllocatedPct":67.84616749173645,"vramReservedGb":60.108570624,"vramReservedPct":70.70141466964319,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.111420928,"vramAllocatedPct":98.93425160086922,"vramReservedGb":84.204847104,"vramReservedPct":99.0441421329814,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 24.19 MiB is free. Including non-PyTorch memory, this process has 79.15 GiB memory in use. Of the allocated memory 78.33 GiB is allocated by PyTorch, and 89.10 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":95253.62532070444,"meanTps":95216.50318212208,"stepMs":344.0079040527344,"jitter":0.0021657430934940414,"achievedTflops":249.8911100669773,"nominalPeakTflops":989.5,"mfuNominalPct":25.25428095674354,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.25428095674354,"vramAllocatedGb":57.68111104,"vramAllocatedPct":38.423141016937905,"vramReservedGb":60.035170304,"vramReservedPct":39.99125143353818,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":98691.95957491407,"meanTps":98651.09497232753,"stepMs":664.0459899902344,"jitter":0.0018313563218644358,"achievedTflops":258.91133539355076,"nominalPeakTflops":989.5,"mfuNominalPct":26.16587522926233,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.16587522926233,"vramAllocatedGb":111.831865344,"vramAllocatedPct":74.49460412300205,"vramReservedGb":116.534542336,"vramReservedPct":77.62720018370243,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.668212736,"vramAllocatedPct":99.03241459288388,"vramReservedGb":148.878917632,"vramReservedPct":99.17277152751979,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 435.44 MiB is free. Including non-PyTorch memory, this process has 139.38 GiB memory in use. Of the allocated memory 138.46 GiB is allocated by PyTorch, and 200.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":95082.9434201925,"meanTps":95052.45096149173,"stepMs":344.62542724609375,"jitter":0.0013602146823173866,"achievedTflops":249.4433382426119,"nominalPeakTflops":989.5,"mfuNominalPct":25.20902862482182,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.20902862482182,"vramAllocatedGb":57.68111104,"vramAllocatedPct":38.423141016937905,"vramReservedGb":60.035170304,"vramReservedPct":39.99125143353818,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":98609.8611989627,"meanTps":98594.43927535792,"stepMs":664.5988464355469,"jitter":0.001769514704920084,"achievedTflops":258.69595614439237,"nominalPeakTflops":989.5,"mfuNominalPct":26.14410875638124,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.14410875638124,"vramAllocatedGb":111.831865344,"vramAllocatedPct":74.49460412300205,"vramReservedGb":116.534542336,"vramReservedPct":77.62720018370243,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.668212736,"vramAllocatedPct":99.03241459288388,"vramReservedGb":148.878917632,"vramReservedPct":99.17277152751979,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 435.44 MiB is free. Including non-PyTorch memory, this process has 139.38 GiB memory in use. Of the allocated memory 138.46 GiB is allocated by PyTorch, and 200.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":32775.903906026164,"meanTps":32764.58065684594,"stepMs":124.96985626220703,"jitter":0.004412124750054166,"achievedTflops":85.9852523507594,"nominalPeakTflops":364.2,"mfuNominalPct":23.609349904107468,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":19.047156685922452,"vramAllocatedGb":10.249098752,"vramAllocatedPct":20.144919284147367,"vramReservedGb":10.311696384,"vramReservedPct":20.26795685794113,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":33094.10948413049,"meanTps":33105.10703788117,"stepMs":247.5364990234375,"jitter":0.0027906788216798444,"achievedTflops":86.82004204904408,"nominalPeakTflops":364.2,"mfuNominalPct":23.838561792708425,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":19.232076422136746,"vramAllocatedGb":17.017975808,"vramAllocatedPct":33.44935562893604,"vramReservedGb":17.620271104,"vramReservedPct":34.63318558479182,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":31718.25994915232,"meanTps":31717.34438293384,"stepMs":516.5478820800781,"jitter":0.0018083328129382063,"achievedTflops":83.21059866646138,"nominalPeakTflops":364.2,"mfuNominalPct":22.847501006716467,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":18.432524966771258,"vramAllocatedGb":30.555664384,"vramAllocatedPct":60.058099505486794,"vramReservedGb":31.694258176,"vramReservedPct":62.29604067400128,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":20582.531856059544,"meanTps":20565.87364705257,"stepMs":398.0074005126953,"jitter":0.0013670552154318846,"achievedTflops":53.9968081653861,"nominalPeakTflops":154.8,"mfuNominalPct":34.881659021567245,"configuredPeakTflops":180.6,"mfuConfiguredPct":29.89856487562907,"vramAllocatedGb":17.017975808,"vramAllocatedPct":33.34639369487721,"vramReservedGb":17.788043264,"vramReservedPct":34.855325946815015,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":22100.915638387367,"meanTps":22095.313903307477,"stepMs":741.3267517089844,"jitter":0.0006059961974814953,"achievedTflops":57.98018000657441,"nominalPeakTflops":154.8,"mfuNominalPct":37.45489664507391,"configuredPeakTflops":180.6,"mfuConfiguredPct":32.104197124349064,"vramAllocatedGb":30.555664384,"vramAllocatedPct":59.87323202553948,"vramReservedGb":31.736201216,"vramReservedPct":62.18647106262103,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.24448,"vramAllocatedPct":96.49360393423277,"vramReservedGb":49.56618752,"vramReservedPct":97.12398358323189,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.23 GiB is free. Process 1415479 has 46.29 GiB memory in use. Of the allocated memory 45.86 GiB is allocated by PyTorch, and 106.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":20618.85462720193,"meanTps":20616.081837492005,"stepMs":397.30625915527344,"jitter":0.0004035918308656897,"achievedTflops":54.09209837162151,"nominalPeakTflops":154.8,"mfuNominalPct":34.943216002339476,"configuredPeakTflops":180.6,"mfuConfiguredPct":29.95132800200527,"vramAllocatedGb":17.017975808,"vramAllocatedPct":33.34639369487721,"vramReservedGb":17.788043264,"vramReservedPct":34.855325946815015,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":22132.4737217773,"meanTps":22132.258668828053,"stepMs":740.2697143554688,"jitter":0.0006657812840234622,"achievedTflops":58.06297039343208,"nominalPeakTflops":154.8,"mfuNominalPct":37.50837880712666,"configuredPeakTflops":180.6,"mfuConfiguredPct":32.15003897753714,"vramAllocatedGb":30.555664384,"vramAllocatedPct":59.87323202553948,"vramReservedGb":31.736201216,"vramReservedPct":62.18647106262103,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.24448,"vramAllocatedPct":96.49360393423277,"vramReservedGb":49.56618752,"vramReservedPct":97.12398358323189,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.23 GiB is free. Process 1449495 has 46.29 GiB memory in use. Of the allocated memory 45.86 GiB is allocated by PyTorch, and 106.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":61365.09144111193,"meanTps":61357.501003184916,"stepMs":133.49609375,"jitter":0.0004062033878931968,"achievedTflops":160.98695212861242,"nominalPeakTflops":468.0,"mfuNominalPct":34.39892139500265,"configuredPeakTflops":468.0,"mfuConfiguredPct":34.39892139500265,"vramAllocatedGb":17.017975808,"vramAllocatedPct":16.688552057231398,"vramReservedGb":17.620271104,"vramReservedPct":17.279188482768948,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":63722.778905228675,"meanTps":63721.116503823025,"stepMs":257.11370849609375,"jitter":0.00024706767941800586,"achievedTflops":167.17217747426724,"nominalPeakTflops":468.0,"mfuNominalPct":35.72055074236479,"configuredPeakTflops":468.0,"mfuConfiguredPct":35.72055074236479,"vramAllocatedGb":30.555664384,"vramAllocatedPct":29.96418619163637,"vramReservedGb":31.715229696,"vramReservedPct":31.101305335029135,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":58371.63369788135,"meanTps":58368.93085156372,"stepMs":561.3685607910156,"jitter":0.00020826605164196103,"achievedTflops":153.13382868185698,"nominalPeakTflops":468.0,"mfuNominalPct":32.72090356449935,"configuredPeakTflops":468.0,"mfuConfiguredPct":32.72090356449935,"vramAllocatedGb":57.631041536,"vramAllocatedPct":56.515454460446314,"vramReservedGb":59.972255744,"vramReservedPct":58.81134885048127,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":60963.47517676136,"meanTps":60915.87861295891,"stepMs":134.37554168701172,"jitter":0.0005516965632689866,"achievedTflops":159.93334042846325,"nominalPeakTflops":468.0,"mfuNominalPct":34.173790689842576,"configuredPeakTflops":468.0,"mfuConfiguredPct":34.173790689842576,"vramAllocatedGb":17.049906176,"vramAllocatedPct":16.7198643363524,"vramReservedGb":18.113101824,"vramReservedPct":17.762479281799024,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":63457.80341600907,"meanTps":63438.616123804335,"stepMs":258.18731689453125,"jitter":0.0006267543918748807,"achievedTflops":166.477033127596,"nominalPeakTflops":468.0,"mfuNominalPct":35.572015625554705,"configuredPeakTflops":468.0,"mfuConfiguredPct":35.572015625554705,"vramAllocatedGb":30.584666112,"vramAllocatedPct":29.992626521610877,"vramReservedGb":33.971765248,"vramReservedPct":33.3141602276094,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":58280.01378431199,"meanTps":58241.256654273566,"stepMs":562.2510681152344,"jitter":0.0008028672519625791,"achievedTflops":152.89347035608193,"nominalPeakTflops":468.0,"mfuNominalPct":32.66954494788075,"configuredPeakTflops":468.0,"mfuConfiguredPct":32.66954494788075,"vramAllocatedGb":57.681934336,"vramAllocatedPct":56.56536210820187,"vramReservedGb":60.110667776,"vramReservedPct":58.94708158552801,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":33561.57851862581,"meanTps":33568.40201500483,"stepMs":488.1772766113281,"jitter":0.002057870016446627,"achievedTflops":126.47593121353256,"nominalPeakTflops":312.0,"mfuNominalPct":40.5371574402348,"configuredPeakTflops":312.0,"mfuConfiguredPct":40.5371574402348,"vramAllocatedGb":30.569935872,"vramAllocatedPct":35.975533248779115,"vramReservedGb":31.748784128,"vramReservedPct":37.36283398786368,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":35229.17524846068,"meanTps":35248.97746328509,"stepMs":930.1381530761719,"jitter":0.0020605957780378712,"achievedTflops":132.76022589226636,"nominalPeakTflops":312.0,"mfuNominalPct":42.55135445264947,"configuredPeakTflops":312.0,"mfuConfiguredPct":42.55135445264947,"vramAllocatedGb":57.645953024,"vramAllocatedPct":67.83932777470993,"vramReservedGb":60.056141824,"vramReservedPct":70.67570360726945,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.251176448,"vramAllocatedPct":99.14907941727004,"vramReservedGb":84.404076544,"vramReservedPct":99.32901641518922,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 58.75 MiB is free. Process 764857 has 79.07 GiB memory in use. Of the allocated memory 78.44 GiB is allocated by PyTorch, and 129.88 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":33511.367797919185,"meanTps":33498.40027268889,"stepMs":488.9087219238281,"jitter":0.0012126447444630681,"achievedTflops":126.28671342525875,"nominalPeakTflops":312.0,"mfuNominalPct":40.47651071322396,"configuredPeakTflops":312.0,"mfuConfiguredPct":40.47651071322396,"vramAllocatedGb":30.569935872,"vramAllocatedPct":35.975533248779115,"vramReservedGb":31.748784128,"vramReservedPct":37.36283398786368,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":35205.753465584035,"meanTps":35204.21124194449,"stepMs":930.7569580078125,"jitter":0.001436951794292306,"achievedTflops":132.67196151583482,"nominalPeakTflops":312.0,"mfuNominalPct":42.5230645884086,"configuredPeakTflops":312.0,"mfuConfiguredPct":42.5230645884086,"vramAllocatedGb":57.645953024,"vramAllocatedPct":67.83932777470993,"vramReservedGb":60.056141824,"vramReservedPct":70.67570360726945,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.251176448,"vramAllocatedPct":99.14907941727004,"vramReservedGb":84.404076544,"vramReservedPct":99.32901641518922,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 58.75 MiB is free. Process 852916 has 79.07 GiB memory in use. Of the allocated memory 78.44 GiB is allocated by PyTorch, and 129.88 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":91566.92467487525,"meanTps":91560.0395023929,"stepMs":357.8584747314453,"jitter":0.0007843945064884667,"achievedTflops":345.0675617115886,"nominalPeakTflops":2250.0,"mfuNominalPct":15.336336076070607,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.336336076070607,"vramAllocatedGb":57.645953024,"vramAllocatedPct":30.101852552133668,"vramReservedGb":60.070821888,"vramReservedPct":31.36808272395507,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":96876.08878726182,"meanTps":96818.33406870744,"stepMs":676.4930419921875,"jitter":0.000665745689466171,"achievedTflops":365.07500786633096,"nominalPeakTflops":2250.0,"mfuNominalPct":16.225555905170268,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.225555905170268,"vramAllocatedGb":111.797987328,"vramAllocatedPct":58.379233122777286,"vramReservedGb":116.48630784,"vramReservedPct":60.827403815880615,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.5517184,"vramAllocatedPct":97.41451092474806,"vramReservedGb":187.315519488,"vramReservedPct":97.81335640346819,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.67 GiB is free. Including non-PyTorch memory, this process has 174.66 GiB memory in use. Of the allocated memory 173.74 GiB is allocated by PyTorch, and 108.42 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":91708.08856126443,"meanTps":91719.85959318749,"stepMs":357.30763244628906,"jitter":0.0010740636917204538,"achievedTflops":345.5995341268577,"nominalPeakTflops":2250.0,"mfuNominalPct":15.359979294527008,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.359979294527008,"vramAllocatedGb":57.645953024,"vramAllocatedPct":30.101852552133668,"vramReservedGb":60.070821888,"vramReservedPct":31.36808272395507,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":96880.6515121166,"meanTps":96875.30390205959,"stepMs":676.461181640625,"jitter":0.00035984332959792983,"achievedTflops":365.0922023756583,"nominalPeakTflops":2250.0,"mfuNominalPct":16.226320105584815,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.226320105584815,"vramAllocatedGb":111.797987328,"vramAllocatedPct":58.379233122777286,"vramReservedGb":116.48630784,"vramReservedPct":60.827403815880615,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.5517184,"vramAllocatedPct":97.41451092474806,"vramReservedGb":187.315519488,"vramReservedPct":97.81335640346819,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.67 GiB is free. Including non-PyTorch memory, this process has 174.66 GiB memory in use. Of the allocated memory 173.74 GiB is allocated by PyTorch, and 108.42 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":97655.9771324567,"meanTps":97473.52101994018,"stepMs":671.0905151367188,"jitter":0.0004523251667303766,"achievedTflops":368.01399670579696,"nominalPeakTflops":2250.0,"mfuNominalPct":16.356177631368755,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.356177631368755,"vramAllocatedGb":111.797987328,"vramAllocatedPct":38.89591066126167,"vramReservedGb":116.467433472,"vramReservedPct":40.520469066966605,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":16,"tokensPerStep":131072,"status":"complete","stable":true,"tps":101359.17902734944,"meanTps":101361.13430897184,"stepMs":1293.1438598632812,"jitter":0.00011602237215890654,"achievedTflops":381.9694162301899,"nominalPeakTflops":2250.0,"mfuNominalPct":16.97641849911955,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.97641849911955,"vramAllocatedGb":220.102055936,"vramAllocatedPct":76.57624353227098,"vramReservedGb":229.279531008,"vramReservedPct":79.76920128605575,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.070432256,"vramAllocatedPct":99.52746236131135,"vramReservedGb":286.206722048,"vramReservedPct":99.57487927551858,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 311.62 MiB is free. Including non-PyTorch memory, this process has 267.37 GiB memory in use. Of the allocated memory 266.42 GiB is allocated by PyTorch, and 129.98 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":97712.3153717999,"meanTps":97708.57287851075,"stepMs":670.7035827636719,"jitter":0.00042418564525768603,"achievedTflops":368.22630588785483,"nominalPeakTflops":2250.0,"mfuNominalPct":16.36561359501577,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.36561359501577,"vramAllocatedGb":111.797987328,"vramAllocatedPct":38.89591066126167,"vramReservedGb":116.467433472,"vramReservedPct":40.520469066966605,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":16,"tokensPerStep":131072,"status":"complete","stable":true,"tps":101399.70661618882,"meanTps":101330.88062448117,"stepMs":1292.6270141601562,"jitter":8.87667015087683e-05,"achievedTflops":382.1221433891777,"nominalPeakTflops":2250.0,"mfuNominalPct":16.983206372852344,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.983206372852344,"vramAllocatedGb":220.102055936,"vramAllocatedPct":76.57624353227098,"vramReservedGb":229.279531008,"vramReservedPct":79.76920128605575,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.070432256,"vramAllocatedPct":99.52746236131135,"vramReservedGb":286.206722048,"vramReservedPct":99.57487927551858,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 311.62 MiB is free. Including non-PyTorch memory, this process has 267.37 GiB memory in use. Of the allocated memory 266.42 GiB is allocated by PyTorch, and 129.98 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":30020.24096784372,"meanTps":30016.797290046976,"stepMs":272.88255310058594,"jitter":0.00047826525832477755,"achievedTflops":113.13049323813921,"nominalPeakTflops":165.2,"mfuNominalPct":68.48092811025376,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":55.04687044269202,"vramAllocatedGb":17.03186176,"vramAllocatedPct":67.4234176248466,"vramReservedGb":17.108566016,"vramReservedPct":67.72706400002076,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.33770752,"vramAllocatedPct":96.34480606260685,"vramReservedGb":24.36890624,"vramReservedPct":96.46831131162554,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 371.69 MiB is free. Including non-PyTorch memory, this process has 23.15 GiB memory in use. Of the allocated memory 22.67 GiB is allocated by PyTorch, and 29.75 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":29833.70970633348,"meanTps":29822.498912189163,"stepMs":274.5887145996094,"jitter":0.006291275092598427,"achievedTflops":112.42755505581115,"nominalPeakTflops":165.2,"mfuNominalPct":68.05542073596317,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":54.70483580689863,"vramAllocatedGb":17.03186176,"vramAllocatedPct":67.4234176248466,"vramReservedGb":17.108566016,"vramReservedPct":67.72706400002076,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.33770752,"vramAllocatedPct":96.34480606260685,"vramReservedGb":24.36890624,"vramReservedPct":96.46831131162554,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 371.69 MiB is free. Including non-PyTorch memory, this process has 23.15 GiB memory in use. Of the allocated memory 22.67 GiB is allocated by PyTorch, and 29.75 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":42674.063375172715,"meanTps":42665.083032685085,"stepMs":191.9667205810547,"jitter":0.0006371392012117579,"achievedTflops":160.8160921586255,"nominalPeakTflops":209.5,"mfuNominalPct":76.76185783227947,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":59.794754628574985,"vramAllocatedGb":17.03186176,"vramAllocatedPct":50.586395988971205,"vramReservedGb":17.108566016,"vramReservedPct":50.81421558513351,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":43052.395976800646,"meanTps":43048.826978789635,"stepMs":380.5595397949219,"jitter":0.0003336100415591601,"achievedTflops":162.2418286767328,"nominalPeakTflops":209.5,"mfuNominalPct":77.44240032302282,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":60.324873002432334,"vramAllocatedGb":30.569935872,"vramAllocatedPct":90.79588028422606,"vramReservedGb":31.72990976,"vramReservedPct":94.24112304524026,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":42654.19032196363,"meanTps":42655.206326164654,"stepMs":192.05615997314453,"jitter":0.00031892186070217176,"achievedTflops":160.74120107716814,"nominalPeakTflops":209.5,"mfuNominalPct":76.72611029936427,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":59.76690857299993,"vramAllocatedGb":17.03186176,"vramAllocatedPct":50.586395988971205,"vramReservedGb":17.108566016,"vramReservedPct":50.81421558513351,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":43032.194939591755,"meanTps":43034.36769409646,"stepMs":380.7381896972656,"jitter":0.0003126572245062912,"achievedTflops":162.16570159614704,"nominalPeakTflops":209.5,"mfuNominalPct":77.40606281439,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":60.296567376775634,"vramAllocatedGb":30.569935872,"vramAllocatedPct":90.79588028422606,"vramReservedGb":31.72990976,"vramReservedPct":94.24112304524026,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":67275.47592083519,"meanTps":67248.991564335,"stepMs":243.53599548339844,"jitter":0.0009310156246328798,"achievedTflops":253.52587215762495,"nominalPeakTflops":989.5,"mfuNominalPct":25.621614164489635,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.621614164489635,"vramAllocatedGb":30.620005376,"vramAllocatedPct":36.01612340471947,"vramReservedGb":31.769755648,"vramReservedPct":37.36849245797309,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":71676.74214026412,"meanTps":71673.93672874398,"stepMs":457.16363525390625,"jitter":0.0003017092204461608,"achievedTflops":270.1119288388388,"nominalPeakTflops":989.5,"mfuNominalPct":27.297819993818983,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.297819993818983,"vramAllocatedGb":57.696022528,"vramAllocatedPct":67.86370680909977,"vramReservedGb":60.119056384,"vramReservedPct":70.7137483195402,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.125085184,"vramAllocatedPct":98.95032388839127,"vramReservedGb":84.215332864,"vramReservedPct":99.05647578287842,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 14.19 MiB is free. Including non-PyTorch memory, this process has 79.16 GiB memory in use. Of the allocated memory 78.35 GiB is allocated by PyTorch, and 86.07 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":67280.48389933648,"meanTps":67184.13458498397,"stepMs":243.5178680419922,"jitter":0.0006665072412523269,"achievedTflops":253.54474459367844,"nominalPeakTflops":989.5,"mfuNominalPct":25.623521434429353,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.623521434429353,"vramAllocatedGb":30.620005376,"vramAllocatedPct":36.01612340471947,"vramReservedGb":31.769755648,"vramReservedPct":37.36849245797309,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":71668.0445724901,"meanTps":71657.50052644055,"stepMs":457.2191162109375,"jitter":0.0002740139269675205,"achievedTflops":270.0791522820716,"nominalPeakTflops":989.5,"mfuNominalPct":27.29450755756156,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.29450755756156,"vramAllocatedGb":57.696022528,"vramAllocatedPct":67.86370680909977,"vramReservedGb":60.119056384,"vramReservedPct":70.7137483195402,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.125085184,"vramAllocatedPct":98.95032388839127,"vramReservedGb":84.215332864,"vramReservedPct":99.05647578287842,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 14.19 MiB is free. Including non-PyTorch memory, this process has 79.16 GiB memory in use. Of the allocated memory 78.35 GiB is allocated by PyTorch, and 86.07 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":77983.41763958757,"meanTps":78014.6646453346,"stepMs":420.19189453125,"jitter":0.00447509159807705,"achievedTflops":293.878470576315,"nominalPeakTflops":989.5,"mfuNominalPct":29.69969384298282,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.69969384298282,"vramAllocatedGb":57.696022528,"vramAllocatedPct":38.43307401226109,"vramReservedGb":60.049850368,"vramReservedPct":40.00103027030189,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":80460.89731725027,"meanTps":80451.96403115339,"stepMs":814.5074462890625,"jitter":0.0008560710728975659,"achievedTflops":303.2147879703582,"nominalPeakTflops":989.5,"mfuNominalPct":30.64323274081437,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.64323274081437,"vramAllocatedGb":111.848056832,"vramAllocatedPct":74.50538976522498,"vramReservedGb":116.5492224,"vramReservedPct":77.63697902046614,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.681844224,"vramAllocatedPct":99.04149494130733,"vramReservedGb":148.889403392,"vramReservedPct":99.17975641092244,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 425.44 MiB is free. Including non-PyTorch memory, this process has 139.39 GiB memory in use. Of the allocated memory 138.47 GiB is allocated by PyTorch, and 197.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":77842.87885203298,"meanTps":77806.3261919722,"stepMs":420.9505157470703,"jitter":0.001807931269234995,"achievedTflops":293.3488538296616,"nominalPeakTflops":989.5,"mfuNominalPct":29.64617016974852,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.64617016974852,"vramAllocatedGb":57.696022528,"vramAllocatedPct":38.43307401226109,"vramReservedGb":60.049850368,"vramReservedPct":40.00103027030189,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":80355.48895405767,"meanTps":80359.0410168279,"stepMs":815.5758972167969,"jitter":0.0017242675761243557,"achievedTflops":302.81755930946315,"nominalPeakTflops":989.5,"mfuNominalPct":30.6030883587128,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.6030883587128,"vramAllocatedGb":111.848056832,"vramAllocatedPct":74.50538976522498,"vramReservedGb":116.5492224,"vramReservedPct":77.63697902046614,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.681844224,"vramAllocatedPct":99.04149494130733,"vramReservedGb":148.889403392,"vramReservedPct":99.17975641092244,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 425.44 MiB is free. Including non-PyTorch memory, this process has 139.39 GiB memory in use. Of the allocated memory 138.47 GiB is allocated by PyTorch, and 197.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":29311.729525597613,"meanTps":29326.012499796256,"stepMs":279.4785614013672,"jitter":0.001941409639173206,"achievedTflops":110.46048639135789,"nominalPeakTflops":364.2,"mfuNominalPct":30.32962284221798,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":24.468826157731414,"vramAllocatedGb":17.03186176,"vramAllocatedPct":33.47664889529948,"vramReservedGb":17.108566016,"vramReservedPct":33.6274134730697,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":27983.35231307526,"meanTps":27977.943493582312,"stepMs":585.490966796875,"jitter":0.003094042230516506,"achievedTflops":105.45453159506124,"nominalPeakTflops":364.2,"mfuNominalPct":28.955115759215058,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":23.359924308159005,"vramAllocatedGb":30.569935872,"vramAllocatedPct":60.0861505547333,"vramReservedGb":31.70893824,"vramReservedPct":62.324894791960524,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":18151.20369556757,"meanTps":18153.626944928885,"stepMs":451.31993103027344,"jitter":0.0011586030328318596,"achievedTflops":68.40233658167696,"nominalPeakTflops":154.8,"mfuNominalPct":44.18755593131586,"configuredPeakTflops":180.6,"mfuConfiguredPct":37.87504794112789,"vramAllocatedGb":17.03186176,"vramAllocatedPct":33.37360294863596,"vramReservedGb":17.276338176,"vramReservedPct":33.85264974650578,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19330.940809261727,"meanTps":19331.906478036424,"stepMs":847.5531616210938,"jitter":0.0004734049525464322,"achievedTflops":72.8481450515863,"nominalPeakTflops":154.8,"mfuNominalPct":47.059525227122926,"configuredPeakTflops":180.6,"mfuConfiguredPct":40.33673590896252,"vramAllocatedGb":30.569935872,"vramAllocatedPct":59.901196729616444,"vramReservedGb":31.75088128,"vramReservedPct":62.21523636344958,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.258367488,"vramAllocatedPct":96.52081619775373,"vramReservedGb":49.572478976,"vramReservedPct":97.13631156930126,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.23 GiB is free. Process 1415479 has 46.29 GiB memory in use. Of the allocated memory 45.88 GiB is allocated by PyTorch, and 99.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":18228.48541056464,"meanTps":18235.6462887082,"stepMs":449.40650939941406,"jitter":0.0013137475929303325,"achievedTflops":68.69357070419028,"nominalPeakTflops":154.8,"mfuNominalPct":44.37569166937356,"configuredPeakTflops":180.6,"mfuConfiguredPct":38.03630714517734,"vramAllocatedGb":17.03186176,"vramAllocatedPct":33.37360294863596,"vramReservedGb":17.276338176,"vramReservedPct":33.85264974650578,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19317.221104068685,"meanTps":19316.356841337216,"stepMs":848.1551208496094,"jitter":0.0007163249940712096,"achievedTflops":72.79644270125429,"nominalPeakTflops":154.8,"mfuNominalPct":47.02612577600406,"configuredPeakTflops":180.6,"mfuConfiguredPct":40.308107808003484,"vramAllocatedGb":30.569935872,"vramAllocatedPct":59.901196729616444,"vramReservedGb":31.75088128,"vramReservedPct":62.21523636344958,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.258367488,"vramAllocatedPct":96.52081619775373,"vramReservedGb":49.572478976,"vramReservedPct":97.13631156930126,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.23 GiB is free. Process 1449495 has 46.29 GiB memory in use. Of the allocated memory 45.88 GiB is allocated by PyTorch, and 99.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":54216.93979841496,"meanTps":54205.90403843749,"stepMs":151.09668731689453,"jitter":0.0005573446398288877,"achievedTflops":204.31512018265272,"nominalPeakTflops":468.0,"mfuNominalPct":43.65707696210528,"configuredPeakTflops":468.0,"mfuConfiguredPct":43.65707696210528,"vramAllocatedGb":17.03186176,"vramAllocatedPct":16.70216921331569,"vramReservedGb":17.108566016,"vramReservedPct":16.777388674414315,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":56325.99209253112,"meanTps":56326.52128968033,"stepMs":290.8781433105469,"jitter":0.00031397258247525703,"achievedTflops":212.26302861396627,"nominalPeakTflops":468.0,"mfuNominalPct":45.35534799443724,"configuredPeakTflops":468.0,"mfuConfiguredPct":45.35534799443724,"vramAllocatedGb":30.569935872,"vramAllocatedPct":29.978181420746413,"vramReservedGb":31.72990976,"vramReservedPct":31.11570123117046,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":52424.49955277106,"meanTps":52412.704079051044,"stepMs":625.05126953125,"jitter":0.00025986803816874395,"achievedTflops":197.56035597850143,"nominalPeakTflops":468.0,"mfuNominalPct":42.21375127745757,"configuredPeakTflops":468.0,"mfuConfiguredPct":42.21375127745757,"vramAllocatedGb":57.645953024,"vramAllocatedPct":56.530077300820885,"vramReservedGb":59.986935808,"vramReservedPct":58.82574474662259,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":53827.80515685078,"meanTps":53781.15093721438,"stepMs":152.18900299072266,"jitter":0.0022258639823294064,"achievedTflops":202.84867645945417,"nominalPeakTflops":468.0,"mfuNominalPct":43.34373428620816,"configuredPeakTflops":468.0,"mfuConfiguredPct":43.34373428620816,"vramAllocatedGb":17.063507968,"vramAllocatedPct":16.73320283303524,"vramReservedGb":17.597202432,"vramReservedPct":17.256566360261157,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":56141.09128341681,"meanTps":56138.7105113709,"stepMs":291.8361511230469,"jitter":0.00030015136642313144,"achievedTflops":211.56623474886575,"nominalPeakTflops":468.0,"mfuNominalPct":45.20646041642431,"configuredPeakTflops":468.0,"mfuConfiguredPct":45.20646041642431,"vramAllocatedGb":30.5982976,"vramAllocatedPct":30.005994139456387,"vramReservedGb":33.992736768,"vramReservedPct":33.33472579352557,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":52346.33662799663,"meanTps":52337.82673712142,"stepMs":625.9845886230469,"jitter":0.00035127455220323566,"achievedTflops":197.2658010399805,"nominalPeakTflops":468.0,"mfuNominalPct":42.150812188030024,"configuredPeakTflops":468.0,"mfuConfiguredPct":42.150812188030024,"vramAllocatedGb":57.695565824,"vramAllocatedPct":56.578729726047385,"vramReservedGb":60.121153536,"vramReservedPct":58.9573643684861,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":26129.493181555696,"meanTps":26094.569100787256,"stepMs":627.0309143066406,"jitter":0.0020899872966501573,"achievedTflops":158.3072025472999,"nominalPeakTflops":312.0,"mfuNominalPct":50.73948799592946,"configuredPeakTflops":312.0,"mfuConfiguredPct":50.73948799592946,"vramAllocatedGb":30.597387776,"vramAllocatedPct":36.00783939718681,"vramReservedGb":30.729568256,"vramReservedPct":36.16339298660192,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":26947.032985477428,"meanTps":26937.256679458183,"stepMs":1216.01513671875,"jitter":0.0012358229645129154,"achievedTflops":163.26031964110078,"nominalPeakTflops":312.0,"mfuNominalPct":52.32702552599384,"configuredPeakTflops":312.0,"mfuConfiguredPct":52.32702552599384,"vramAllocatedGb":57.673856,"vramAllocatedPct":67.87216475693427,"vramReservedGb":60.087599104,"vramReservedPct":70.71272339125902,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.27850496,"vramAllocatedPct":99.18124035461096,"vramReservedGb":84.414562304,"vramReservedPct":99.34135634318574,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 28.75 MiB is free. Process 925444 has 79.10 GiB memory in use. Of the allocated memory 78.47 GiB is allocated by PyTorch, and 133.88 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":26058.770103083574,"meanTps":26052.45991904976,"stepMs":628.732666015625,"jitter":0.0016741425137139945,"achievedTflops":157.87872226141522,"nominalPeakTflops":312.0,"mfuNominalPct":50.60215457096642,"configuredPeakTflops":312.0,"mfuConfiguredPct":50.60215457096642,"vramAllocatedGb":30.597387776,"vramAllocatedPct":36.00783939718681,"vramReservedGb":30.729568256,"vramReservedPct":36.16339298660192,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":26892.421230966218,"meanTps":26893.79477296766,"stepMs":1218.4845581054688,"jitter":0.0005566159716977621,"achievedTflops":162.9294508399802,"nominalPeakTflops":312.0,"mfuNominalPct":52.220977833326984,"configuredPeakTflops":312.0,"mfuConfiguredPct":52.220977833326984,"vramAllocatedGb":57.673856,"vramAllocatedPct":67.87216475693427,"vramReservedGb":60.087599104,"vramReservedPct":70.71272339125902,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.27850496,"vramAllocatedPct":99.18124035461096,"vramReservedGb":84.414562304,"vramReservedPct":99.34135634318574,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 28.75 MiB is free. Process 937161 has 79.10 GiB memory in use. Of the allocated memory 78.47 GiB is allocated by PyTorch, and 133.88 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":65311.726133788274,"meanTps":65291.082723808446,"stepMs":501.7169494628906,"jitter":0.0007274454401910042,"achievedTflops":395.6952622821526,"nominalPeakTflops":2250.0,"mfuNominalPct":17.586456101429004,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.586456101429004,"vramAllocatedGb":57.673856,"vramAllocatedPct":30.116423067933244,"vramReservedGb":60.09389056,"vramReservedPct":31.38012883867241,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":68328.73041834016,"meanTps":68300.94493764888,"stepMs":959.1280212402344,"jitter":0.0007382170510546297,"achievedTflops":413.97397534566375,"nominalPeakTflops":2250.0,"mfuNominalPct":18.398843348696165,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.398843348696165,"vramAllocatedGb":111.826530304,"vramAllocatedPct":58.39413783698321,"vramReservedGb":116.511473664,"vramReservedPct":60.840545031935896,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.579493376,"vramAllocatedPct":97.42901460086637,"vramReservedGb":187.342782464,"vramReservedPct":97.8275927208614,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.65 GiB is free. Including non-PyTorch memory, this process has 174.69 GiB memory in use. Of the allocated memory 173.77 GiB is allocated by PyTorch, and 107.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":65488.93582306832,"meanTps":65402.43735710658,"stepMs":500.3593292236328,"jitter":0.0006556770221498228,"achievedTflops":396.7688984977224,"nominalPeakTflops":2250.0,"mfuNominalPct":17.63417326656544,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.63417326656544,"vramAllocatedGb":57.673856,"vramAllocatedPct":30.116423067933244,"vramReservedGb":60.09389056,"vramReservedPct":31.38012883867241,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":68316.27517298942,"meanTps":68299.6378784464,"stepMs":959.3028869628906,"jitter":0.00026289678418608227,"achievedTflops":413.8985144465634,"nominalPeakTflops":2250.0,"mfuNominalPct":18.395489530958375,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.395489530958375,"vramAllocatedGb":111.826530304,"vramAllocatedPct":58.39413783698321,"vramReservedGb":116.511473664,"vramReservedPct":60.840545031935896,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.579493376,"vramAllocatedPct":97.42901460086637,"vramReservedGb":187.342782464,"vramReservedPct":97.8275927208614,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.65 GiB is free. Including non-PyTorch memory, this process has 174.69 GiB memory in use. Of the allocated memory 173.77 GiB is allocated by PyTorch, and 107.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":69071.70485575692,"meanTps":69070.07977630837,"stepMs":948.8110961914062,"jitter":0.00012813328846667107,"achievedTflops":418.4753334062413,"nominalPeakTflops":2250.0,"mfuNominalPct":18.59890370694406,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.59890370694406,"vramAllocatedGb":111.826530304,"vramAllocatedPct":38.90584111771295,"vramReservedGb":116.490502144,"vramReservedPct":40.528494944950914,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":8,"tokensPerStep":131072,"status":"complete","stable":true,"tps":71060.76875332784,"meanTps":71059.26834484094,"stepMs":1844.5057983398438,"jitter":0.00010528190857144332,"achievedTflops":430.5262040115139,"nominalPeakTflops":2250.0,"mfuNominalPct":19.134497956067285,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.134497956067285,"vramAllocatedGb":220.131878912,"vramAllocatedPct":76.58661931664665,"vramReservedGb":229.306793984,"vramReservedPct":79.77868641458265,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.097695232,"vramAllocatedPct":99.53694748983825,"vramReservedGb":286.217207808,"vramReservedPct":99.57852740187509,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 301.62 MiB is free. Including non-PyTorch memory, this process has 267.38 GiB memory in use. Of the allocated memory 266.45 GiB is allocated by PyTorch, and 113.98 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":69032.45331152566,"meanTps":69034.78366643804,"stepMs":949.3505859375,"jitter":0.00029720324337669153,"achievedTflops":418.23752541969765,"nominalPeakTflops":2250.0,"mfuNominalPct":18.588334463097674,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.588334463097674,"vramAllocatedGb":111.826530304,"vramAllocatedPct":38.90584111771295,"vramReservedGb":116.490502144,"vramReservedPct":40.528494944950914,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":8,"tokensPerStep":131072,"status":"complete","stable":true,"tps":71045.70640812338,"meanTps":71038.69461674929,"stepMs":1844.8968505859375,"jitter":0.00011443179378293229,"achievedTflops":430.4349478315126,"nominalPeakTflops":2250.0,"mfuNominalPct":19.130442125845004,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.130442125845004,"vramAllocatedGb":220.131878912,"vramAllocatedPct":76.58661931664665,"vramReservedGb":229.306793984,"vramReservedPct":79.77868641458265,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.097695232,"vramAllocatedPct":99.53694748983825,"vramReservedGb":286.217207808,"vramReservedPct":99.57852740187509,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 301.62 MiB is free. Including non-PyTorch memory, this process has 267.38 GiB memory in use. Of the allocated memory 266.45 GiB is allocated by PyTorch, and 113.98 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.432052224,"vramAllocatedPct":96.71828504383154,"vramReservedGb":24.534581248,"vramReservedPct":97.12416299782333,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 23.53 GiB of which 253.69 MiB is free. Including non-PyTorch memory, this process has 23.27 GiB memory in use. Of the allocated memory 22.75 GiB is allocated by PyTorch, and 57.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.432052224,"vramAllocatedPct":96.71828504383154,"vramReservedGb":24.534581248,"vramReservedPct":97.12416299782333,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 23.53 GiB of which 253.69 MiB is free. Including non-PyTorch memory, this process has 23.27 GiB memory in use. Of the allocated memory 22.75 GiB is allocated by PyTorch, and 57.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":34179.944205173284,"meanTps":34182.517492644554,"stepMs":479.3454284667969,"jitter":0.00035825682007686795,"achievedTflops":207.0813740146804,"nominalPeakTflops":209.5,"mfuNominalPct":98.84552458934625,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":76.99714488238072,"vramAllocatedGb":30.597387776,"vramAllocatedPct":90.87741528644506,"vramReservedGb":30.729568256,"vramReservedPct":91.27000502193691,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_extension_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.80088832,"vramAllocatedPct":97.42204045140984,"vramReservedGb":32.971423744,"vramReservedPct":97.92854834879493,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 95.88 MiB is free. Including non-PyTorch memory, this process has 31.25 GiB memory in use. Of the allocated memory 30.55 GiB is allocated by PyTorch, and 122.64 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":34169.26412862494,"meanTps":34168.55268701906,"stepMs":479.49525451660156,"jitter":0.0004282074697778314,"achievedTflops":207.01666808909616,"nominalPeakTflops":209.5,"mfuNominalPct":98.81463870601249,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":76.97308587876118,"vramAllocatedGb":30.597387776,"vramAllocatedPct":90.87741528644506,"vramReservedGb":30.729568256,"vramReservedPct":91.27000502193691,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_extension_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.80088832,"vramAllocatedPct":97.42204045140984,"vramReservedGb":32.971423744,"vramReservedPct":97.92854834879493,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 95.88 MiB is free. Including non-PyTorch memory, this process has 31.25 GiB memory in use. Of the allocated memory 30.55 GiB is allocated by PyTorch, and 122.64 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":52114.0393451738,"meanTps":52104.089249348246,"stepMs":314.387451171875,"jitter":0.0009438726042162531,"achievedTflops":315.73623433300725,"nominalPeakTflops":989.5,"mfuNominalPct":31.90866440960154,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.90866440960154,"vramAllocatedGb":30.64745728,"vramAllocatedPct":36.0484131169523,"vramReservedGb":30.750539776,"vramReservedPct":36.16966168798332,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":54221.58590521928,"meanTps":54220.35434926399,"stepMs":604.3349609375,"jitter":0.0005603871220947612,"achievedTflops":328.50493971281526,"nominalPeakTflops":989.5,"mfuNominalPct":33.199084357030344,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.199084357030344,"vramAllocatedGb":57.723925504,"vramAllocatedPct":67.8965270850806,"vramReservedGb":60.150513664,"vramReservedPct":70.75074926923125,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.09369344,"vramAllocatedPct":98.91340002401208,"vramReservedGb":84.183875584,"vramReservedPct":99.01947483318739,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 44.19 MiB is free. Including non-PyTorch memory, this process has 79.13 GiB memory in use. Of the allocated memory 78.32 GiB is allocated by PyTorch, and 86.00 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":52107.13006640452,"meanTps":52109.95450585552,"stepMs":314.42913818359375,"jitter":0.0008060381838133228,"achievedTflops":315.69437402649527,"nominalPeakTflops":989.5,"mfuNominalPct":31.90443395922135,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.90443395922135,"vramAllocatedGb":30.64745728,"vramAllocatedPct":36.0484131169523,"vramReservedGb":30.750539776,"vramReservedPct":36.16966168798332,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":54186.19208738812,"meanTps":54174.95618341588,"stepMs":604.7297058105469,"jitter":0.0010527054116408116,"achievedTflops":328.2905040079438,"nominalPeakTflops":989.5,"mfuNominalPct":33.17741323981241,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.17741323981241,"vramAllocatedGb":57.723925504,"vramAllocatedPct":67.8965270850806,"vramReservedGb":60.150513664,"vramReservedPct":70.75074926923125,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.09369344,"vramAllocatedPct":98.91340002401208,"vramReservedGb":84.183875584,"vramReservedPct":99.01947483318739,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 44.19 MiB is free. Including non-PyTorch memory, this process has 79.13 GiB memory in use. Of the allocated memory 78.32 GiB is allocated by PyTorch, and 86.00 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":57787.837503722934,"meanTps":57767.2777930779,"stepMs":567.0397338867188,"jitter":0.0003273077018709355,"achievedTflops":350.1113026918516,"nominalPeakTflops":989.5,"mfuNominalPct":35.38264807396176,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.38264807396176,"vramAllocatedGb":57.723925504,"vramAllocatedPct":38.45166103255785,"vramReservedGb":60.07291904,"vramReservedPct":40.016397013787724,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":59094.17578228214,"meanTps":59117.34573335731,"stepMs":1109.0094604492188,"jitter":0.0007065396603227153,"achievedTflops":358.0258365491383,"nominalPeakTflops":989.5,"mfuNominalPct":36.182499903904834,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.182499903904834,"vramAllocatedGb":111.876599808,"vramAllocatedPct":74.5244031089716,"vramReservedGb":116.574388224,"vramReservedPct":77.65374274063251,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.7091072,"vramAllocatedPct":99.05965563815423,"vramReservedGb":148.916666368,"vramReservedPct":99.19791710776933,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 399.44 MiB is free. Including non-PyTorch memory, this process has 139.41 GiB memory in use. Of the allocated memory 138.50 GiB is allocated by PyTorch, and 197.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":57873.84736516894,"meanTps":57858.233902583845,"stepMs":566.197021484375,"jitter":0.003104861386296576,"achievedTflops":350.6323989282918,"nominalPeakTflops":989.5,"mfuNominalPct":35.43531065470357,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.43531065470357,"vramAllocatedGb":57.723925504,"vramAllocatedPct":38.45166103255785,"vramReservedGb":60.07291904,"vramReservedPct":40.016397013787724,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":59073.745447023175,"meanTps":59081.35510347037,"stepMs":1109.3930053710938,"jitter":0.0009958610046766957,"achievedTflops":357.90205805869925,"nominalPeakTflops":989.5,"mfuNominalPct":36.169990708307154,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.169990708307154,"vramAllocatedGb":111.876599808,"vramAllocatedPct":74.5244031089716,"vramReservedGb":116.574388224,"vramReservedPct":77.65374274063251,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.7091072,"vramAllocatedPct":99.05965563815423,"vramReservedGb":148.916666368,"vramReservedPct":99.19791710776933,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 399.44 MiB is free. Including non-PyTorch memory, this process has 139.41 GiB memory in use. Of the allocated memory 138.50 GiB is allocated by PyTorch, and 197.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":15556.525578614433,"meanTps":15554.700893985122,"stepMs":1053.1914672851562,"jitter":0.0007233645711228467,"achievedTflops":94.25020334662848,"nominalPeakTflops":154.8,"mfuNominalPct":60.88514428076775,"configuredPeakTflops":180.6,"mfuConfiguredPct":52.18726652637236,"vramAllocatedGb":30.597387776,"vramAllocatedPct":59.95498820333729,"vramReservedGb":30.729568256,"vramReservedPct":60.213993291520914,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.285758464,"vramAllocatedPct":96.57448828423969,"vramReservedGb":49.538924544,"vramReservedPct":97.07056231026459,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.18 GiB is free. Process 1472317 has 46.34 GiB memory in use. Of the allocated memory 45.90 GiB is allocated by PyTorch, and 121.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":15552.571524847572,"meanTps":15555.813754652754,"stepMs":1053.459228515625,"jitter":0.0008184423109412836,"achievedTflops":94.22624745945515,"nominalPeakTflops":154.8,"mfuNominalPct":60.86966890145681,"configuredPeakTflops":180.6,"mfuConfiguredPct":52.17400191553442,"vramAllocatedGb":30.597387776,"vramAllocatedPct":59.95498820333729,"vramReservedGb":30.729568256,"vramReservedPct":60.213993291520914,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.285758464,"vramAllocatedPct":96.57448828423969,"vramReservedGb":49.538924544,"vramReservedPct":97.07056231026459,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.18 GiB is free. Process 1479887 has 46.34 GiB memory in use. Of the allocated memory 45.90 GiB is allocated by PyTorch, and 121.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":46099.94002132098,"meanTps":46101.19430674102,"stepMs":355.4017639160156,"jitter":0.0004986003887236649,"achievedTflops":279.2994296393441,"nominalPeakTflops":468.0,"mfuNominalPct":59.67936530755216,"configuredPeakTflops":468.0,"mfuConfiguredPct":59.67936530755216,"vramAllocatedGb":30.621104128,"vramAllocatedPct":30.028359192390226,"vramReservedGb":30.960254976,"vramReservedPct":30.36094496204689,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":43609.53582407988,"meanTps":43608.88704996184,"stepMs":751.3952941894531,"jitter":0.00012030981912271688,"achievedTflops":264.211156822955,"nominalPeakTflops":468.0,"mfuNominalPct":56.45537538952029,"configuredPeakTflops":468.0,"mfuConfiguredPct":56.45537538952029,"vramAllocatedGb":57.703856128,"vramAllocatedPct":56.58685955132362,"vramReservedGb":60.140027904,"vramReservedPct":58.97587337781066,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":46095.14479523576,"meanTps":46095.72400873723,"stepMs":355.43873596191406,"jitter":0.0005287809457469215,"achievedTflops":279.27037745597954,"nominalPeakTflops":468.0,"mfuNominalPct":59.67315757606401,"configuredPeakTflops":468.0,"mfuConfiguredPct":59.67315757606401,"vramAllocatedGb":30.621104128,"vramAllocatedPct":30.028359192390226,"vramReservedGb":30.960254976,"vramReservedPct":30.36094496204689,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":43602.158330378115,"meanTps":43597.75604508451,"stepMs":751.5224304199219,"jitter":0.0001091515525917909,"achievedTflops":264.1664598063836,"nominalPeakTflops":468.0,"mfuNominalPct":56.44582474495376,"configuredPeakTflops":468.0,"mfuConfiguredPct":56.44582474495376,"vramAllocatedGb":57.703856128,"vramAllocatedPct":56.58685955132362,"vramReservedGb":60.140027904,"vramReservedPct":58.97587337781066,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":18349.404440871658,"meanTps":18346.793845553904,"stepMs":1785.780029296875,"jitter":0.0007529998749069492,"achievedTflops":195.21461287581107,"nominalPeakTflops":312.0,"mfuNominalPct":62.56878617814458,"configuredPeakTflops":312.0,"mfuConfiguredPct":62.56878617814458,"vramAllocatedGb":57.728439808,"vramAllocatedPct":67.93640046903295,"vramReservedGb":58.07013888,"vramReservedPct":68.33852124472854,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.274441728,"vramAllocatedPct":99.17645863251232,"vramReservedGb":84.406173696,"vramReservedPct":99.33148440078853,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 56.75 MiB is free. Process 949988 has 79.07 GiB memory in use. Of the allocated memory 78.49 GiB is allocated by PyTorch, and 85.63 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":18391.831433038013,"meanTps":18383.660322829088,"stepMs":1781.6605224609375,"jitter":0.0008806496076870442,"achievedTflops":195.66598277601292,"nominalPeakTflops":312.0,"mfuNominalPct":62.713456017952865,"configuredPeakTflops":312.0,"mfuConfiguredPct":62.713456017952865,"vramAllocatedGb":57.728439808,"vramAllocatedPct":67.93640046903295,"vramReservedGb":58.07013888,"vramReservedPct":68.33852124472854,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.274441728,"vramAllocatedPct":99.17645863251232,"vramReservedGb":84.406173696,"vramReservedPct":99.33148440078853,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 56.75 MiB is free. Process 970883 has 79.07 GiB memory in use. Of the allocated memory 78.49 GiB is allocated by PyTorch, and 85.63 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":42259.08513293796,"meanTps":42257.15143732848,"stepMs":775.4072265625,"jitter":0.000409445114210401,"achievedTflops":449.5835802897886,"nominalPeakTflops":2250.0,"mfuNominalPct":19.981492457323938,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.981492457323938,"vramAllocatedGb":57.728439808,"vramAllocatedPct":30.14492591425562,"vramReservedGb":58.11208192,"vramReservedPct":30.34525807431905,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":43440.70936234824,"meanTps":43440.020611217864,"stepMs":1508.6309814453125,"jitter":0.0002916925276556079,"achievedTflops":462.1545777438099,"nominalPeakTflops":2250.0,"mfuNominalPct":20.54020345528044,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.54020345528044,"vramAllocatedGb":111.881696256,"vramAllocatedPct":58.42294467017601,"vramReservedGb":116.7065088,"vramReservedPct":60.94238945636432,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.634275328,"vramAllocatedPct":97.45762091501534,"vramReservedGb":187.728658432,"vramReservedPct":98.02909136704238,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.52 GiB is free. Including non-PyTorch memory, this process has 174.81 GiB memory in use. Of the allocated memory 173.82 GiB is allocated by PyTorch, and 183.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":42055.61104262049,"meanTps":42051.23585871279,"stepMs":779.1588134765625,"jitter":0.0006296475691417136,"achievedTflops":447.41887157133516,"nominalPeakTflops":2250.0,"mfuNominalPct":19.88528318094823,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.88528318094823,"vramAllocatedGb":57.728439808,"vramAllocatedPct":30.14492591425562,"vramReservedGb":58.11208192,"vramReservedPct":30.34525807431905,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":43345.76657496201,"meanTps":43323.7947228533,"stepMs":1511.9354248046875,"jitter":0.0007214429617847894,"achievedTflops":461.1445057523904,"nominalPeakTflops":2250.0,"mfuNominalPct":20.495311366772906,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.495311366772906,"vramAllocatedGb":111.881696256,"vramAllocatedPct":58.42294467017601,"vramReservedGb":116.7065088,"vramReservedPct":60.94238945636432,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":186.634275328,"vramAllocatedPct":97.45762091501534,"vramReservedGb":187.728658432,"vramReservedPct":98.02909136704238,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 178.35 GiB of which 3.52 GiB is free. Including non-PyTorch memory, this process has 174.81 GiB memory in use. Of the allocated memory 173.82 GiB is allocated by PyTorch, and 183.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":43880.13184859137,"meanTps":43874.031769870635,"stepMs":1493.5233154296875,"jitter":0.00017837330341465422,"achievedTflops":466.829480998425,"nominalPeakTflops":2250.0,"mfuNominalPct":20.747976933263335,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.747976933263335,"vramAllocatedGb":111.881696256,"vramAllocatedPct":38.92503403872897,"vramReservedGb":116.66456576,"vramReservedPct":40.589053842468886,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":4,"tokensPerStep":131072,"status":"complete","stable":true,"tps":44774.780783297545,"meanTps":44762.98514049762,"stepMs":2927.3621826171875,"jitter":0.00014262813013545626,"achievedTflops":476.3474218128641,"nominalPeakTflops":2250.0,"mfuNominalPct":21.170996525016182,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.170996525016182,"vramAllocatedGb":220.187684864,"vramAllocatedPct":76.60603490162485,"vramReservedGb":230.072254464,"vramReservedPct":80.04499963860748,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.152221184,"vramAllocatedPct":99.55591774689208,"vramReservedGb":286.309482496,"vramReservedPct":99.61063091381233,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 213.62 MiB is free. Including non-PyTorch memory, this process has 267.46 GiB memory in use. Of the allocated memory 266.50 GiB is allocated by PyTorch, and 149.98 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":43900.876769024595,"meanTps":43900.793125638265,"stepMs":1492.8175659179688,"jitter":0.0001365103895307949,"achievedTflops":467.0501808922315,"nominalPeakTflops":2250.0,"mfuNominalPct":20.757785817432513,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.757785817432513,"vramAllocatedGb":111.881696256,"vramAllocatedPct":38.92503403872897,"vramReservedGb":116.66456576,"vramReservedPct":40.589053842468886,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":4,"tokensPerStep":131072,"status":"complete","stable":true,"tps":44778.16795491848,"meanTps":44778.85198677011,"stepMs":2927.1407470703125,"jitter":5.2925407615795344e-05,"achievedTflops":476.383457063973,"nominalPeakTflops":2250.0,"mfuNominalPct":21.172598091732134,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.172598091732134,"vramAllocatedGb":220.187684864,"vramAllocatedPct":76.60603490162485,"vramReservedGb":230.072254464,"vramReservedPct":80.04499963860748,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.152221184,"vramAllocatedPct":99.55591774689208,"vramReservedGb":286.309482496,"vramReservedPct":99.61063091381233,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 213.62 MiB is free. Including non-PyTorch memory, this process has 267.46 GiB memory in use. Of the allocated memory 266.50 GiB is allocated by PyTorch, and 149.98 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.966896128,"vramAllocatedPct":97.91510085966605,"vramReservedGb":33.028046848,"vramReservedPct":98.09672484068003,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 152.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 1.88 MiB is free. Including non-PyTorch memory, this process has 31.35 GiB memory in use. Of the allocated memory 30.70 GiB is allocated by PyTorch, and 58.32 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.966896128,"vramAllocatedPct":97.91510085966605,"vramReservedGb":33.028046848,"vramReservedPct":98.09672484068003,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 152.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 1.88 MiB is free. Including non-PyTorch memory, this process has 31.35 GiB memory in use. Of the allocated memory 30.70 GiB is allocated by PyTorch, and 58.32 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_32k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":36458.85613495764,"meanTps":36453.799526891096,"stepMs":898.7665405273438,"jitter":0.0002640417833387242,"achievedTflops":387.8764299525434,"nominalPeakTflops":989.5,"mfuNominalPct":39.199234962359114,"configuredPeakTflops":989.5,"mfuConfiguredPct":39.199234962359114,"vramAllocatedGb":57.778509312,"vramAllocatedPct":67.96073011642194,"vramReservedGb":58.15402496,"vramReservedPct":68.40242232883978,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_32k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.148350464,"vramAllocatedPct":98.97768917410026,"vramReservedGb":84.217430016,"vramReservedPct":99.05894251285783,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 12.19 MiB is free. Including non-PyTorch memory, this process has 79.16 GiB memory in use. Of the allocated memory 78.37 GiB is allocated by PyTorch, and 65.88 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_32k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":36454.4520094492,"meanTps":36448.81496965027,"stepMs":898.8751220703125,"jitter":0.0002366776916406886,"achievedTflops":387.8295755895609,"nominalPeakTflops":989.5,"mfuNominalPct":39.19449980692884,"configuredPeakTflops":989.5,"mfuConfiguredPct":39.19449980692884,"vramAllocatedGb":57.778509312,"vramAllocatedPct":67.96073011642194,"vramReservedGb":58.15402496,"vramReservedPct":68.40242232883978,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_32k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.148350464,"vramAllocatedPct":98.97768917410026,"vramReservedGb":84.217430016,"vramReservedPct":99.05894251285783,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 12.19 MiB is free. Including non-PyTorch memory, this process has 79.16 GiB memory in use. Of the allocated memory 78.37 GiB is allocated by PyTorch, and 65.88 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":38174.429049292514,"meanTps":38205.8065098479,"stepMs":858.3756408691406,"jitter":0.003550101127898981,"achievedTflops":406.12797067209607,"nominalPeakTflops":989.5,"mfuNominalPct":41.043756510570596,"configuredPeakTflops":989.5,"mfuConfiguredPct":41.043756510570596,"vramAllocatedGb":57.778509312,"vramAllocatedPct":38.48802096589151,"vramReservedGb":58.15402496,"vramReservedPct":38.738163351102585,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":38648.39561774428,"meanTps":38642.877156588125,"stepMs":1695.6978149414062,"jitter":0.0010886436993204993,"achievedTflops":411.1703795674113,"nominalPeakTflops":989.5,"mfuNominalPct":41.55334811191625,"configuredPeakTflops":989.5,"mfuConfiguredPct":41.55334811191625,"vramAllocatedGb":111.93176576,"vramAllocatedPct":74.56115082611525,"vramReservedGb":116.85330944,"vramReservedPct":77.83954063914302,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.763633152,"vramAllocatedPct":99.09597703184801,"vramReservedGb":148.904083456,"vramReservedPct":99.18953524768615,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 411.44 MiB is free. Including non-PyTorch memory, this process has 139.40 GiB memory in use. Of the allocated memory 138.55 GiB is allocated by PyTorch, and 133.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":38127.48279713937,"meanTps":38153.76920038429,"stepMs":859.4325561523438,"jitter":0.0021915845906058987,"achievedTflops":405.628521522693,"nominalPeakTflops":989.5,"mfuNominalPct":40.993281609165535,"configuredPeakTflops":989.5,"mfuConfiguredPct":40.993281609165535,"vramAllocatedGb":57.778509312,"vramAllocatedPct":38.48802096589151,"vramReservedGb":58.15402496,"vramReservedPct":38.738163351102585,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":38541.415331950986,"meanTps":38524.23021391324,"stepMs":1700.4046020507812,"jitter":0.0011094715832115582,"achievedTflops":410.0322437143499,"nominalPeakTflops":989.5,"mfuNominalPct":41.43832680286507,"configuredPeakTflops":989.5,"mfuConfiguredPct":41.43832680286507,"vramAllocatedGb":111.93176576,"vramAllocatedPct":74.56115082611525,"vramReservedGb":116.85330944,"vramReservedPct":77.83954063914302,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.763633152,"vramAllocatedPct":99.09597703184801,"vramReservedGb":148.904083456,"vramReservedPct":99.18953524768615,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 411.44 MiB is free. Including non-PyTorch memory, this process has 139.40 GiB memory in use. Of the allocated memory 138.55 GiB is allocated by PyTorch, and 133.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":47.368800256,"vramAllocatedPct":92.81824583673637,"vramReservedGb":47.565504512,"vramReservedPct":93.2036840131704,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 47.53 GiB of which 3.02 GiB is free. Process 1490581 has 44.50 GiB memory in use. Of the allocated memory 44.12 GiB is allocated by PyTorch, and 67.59 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":47.368800256,"vramAllocatedPct":92.81824583673637,"vramReservedGb":47.565504512,"vramReservedPct":93.2036840131704,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 47.53 GiB of which 3.02 GiB is free. Process 1493539 has 44.50 GiB memory in use. Of the allocated memory 44.12 GiB is allocated by PyTorch, and 67.59 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":32847.5034284795,"meanTps":32845.454650015745,"stepMs":997.5796203613281,"jitter":0.00031476465758538066,"achievedTflops":349.456173707995,"nominalPeakTflops":468.0,"mfuNominalPct":74.67012258717841,"configuredPeakTflops":468.0,"mfuConfiguredPct":74.67012258717841,"vramAllocatedGb":57.764544,"vramAllocatedPct":56.646372663960584,"vramReservedGb":60.349743104,"vramReservedPct":59.181529036972385,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":99.34136576,"vramAllocatedPct":97.41837528896227,"vramReservedGb":99.891544064,"vramReservedPct":97.95790357191747,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 94.97 GiB of which 1.61 GiB is free. Including non-PyTorch memory, this process has 93.35 GiB memory in use. Of the allocated memory 92.52 GiB is allocated by PyTorch, and 188.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":32850.484108918135,"meanTps":32851.417818328606,"stepMs":997.4891052246094,"jitter":0.00018048447023796902,"achievedTflops":349.4878844035552,"nominalPeakTflops":468.0,"mfuNominalPct":74.67689837682803,"configuredPeakTflops":468.0,"mfuConfiguredPct":74.67689837682803,"vramAllocatedGb":57.764544,"vramAllocatedPct":56.646372663960584,"vramReservedGb":60.349743104,"vramReservedPct":59.181529036972385,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"250m","modelLabel":"250M","parameters":246397312,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":99.34136576,"vramAllocatedPct":97.41837528896227,"vramReservedGb":99.891544064,"vramReservedPct":97.95790357191747,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 94.97 GiB of which 1.61 GiB is free. Including non-PyTorch memory, this process has 93.35 GiB memory in use. Of the allocated memory 92.52 GiB is allocated by PyTorch, and 188.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":37323.10578017043,"meanTps":37310.425552558925,"stepMs":438.97740173339844,"jitter":0.002153542701416738,"achievedTflops":84.60577902742253,"nominalPeakTflops":312.0,"mfuNominalPct":27.11723686776363,"configuredPeakTflops":312.0,"mfuConfiguredPct":27.11723686776363,"vramAllocatedGb":36.864123392,"vramAllocatedPct":43.382704573832875,"vramReservedGb":38.025560064,"vramReservedPct":44.749514886580634,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":39124.64428641024,"meanTps":39125.52447823315,"stepMs":837.5283813476562,"jitter":0.0011473834805475706,"achievedTflops":88.68959160363367,"nominalPeakTflops":312.0,"mfuNominalPct":28.426151155010793,"configuredPeakTflops":312.0,"mfuConfiguredPct":28.426151155010793,"vramAllocatedGb":68.831059968,"vramAllocatedPct":81.00226630490111,"vramReservedGb":71.370276864,"vramReservedPct":83.99048591551468,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.681412096,"vramAllocatedPct":98.47856520765013,"vramReservedGb":84.12725248,"vramReservedPct":99.00324231608109,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 522.75 MiB is free. Process 764857 has 78.62 GiB memory in use. Of the allocated memory 77.93 GiB is allocated by PyTorch, and 185.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":37229.05747721365,"meanTps":37247.6236260123,"stepMs":440.0863494873047,"jitter":0.0019044045595549663,"achievedTflops":84.39258589219067,"nominalPeakTflops":312.0,"mfuNominalPct":27.0489057346765,"configuredPeakTflops":312.0,"mfuConfiguredPct":27.0489057346765,"vramAllocatedGb":36.864123392,"vramAllocatedPct":43.382704573832875,"vramReservedGb":38.025560064,"vramReservedPct":44.749514886580634,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":39110.069997268554,"meanTps":39115.33270351006,"stepMs":837.8404846191406,"jitter":0.0008639837166829294,"achievedTflops":88.65655391663448,"nominalPeakTflops":312.0,"mfuNominalPct":28.41556215276746,"configuredPeakTflops":312.0,"mfuConfiguredPct":28.41556215276746,"vramAllocatedGb":68.831059968,"vramAllocatedPct":81.00226630490111,"vramReservedGb":71.370276864,"vramReservedPct":83.99048591551468,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.681412096,"vramAllocatedPct":98.47856520765013,"vramReservedGb":84.12725248,"vramReservedPct":99.00324231608109,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 522.75 MiB is free. Process 852916 has 78.62 GiB memory in use. Of the allocated memory 77.93 GiB is allocated by PyTorch, and 185.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":119654.33024378578,"meanTps":119699.02147213905,"stepMs":273.85552978515625,"jitter":0.008116574705322153,"achievedTflops":271.2380873099397,"nominalPeakTflops":2250.0,"mfuNominalPct":12.055026102663987,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.055026102663987,"vramAllocatedGb":68.831059968,"vramAllocatedPct":35.942547732729565,"vramReservedGb":71.267516416,"vramReservedPct":37.21482876721705,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":130305.42963650425,"meanTps":130320.49798745237,"stepMs":502.9414367675781,"jitter":0.00085265500775619,"achievedTflops":295.38250248608045,"nominalPeakTflops":2250.0,"mfuNominalPct":13.128111221603575,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.128111221603575,"vramAllocatedGb":132.764933632,"vramAllocatedPct":69.32785818668671,"vramReservedGb":137.63608576,"vramReservedPct":71.87150080900611,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.409678848,"vramAllocatedPct":99.42908003958792,"vramReservedGb":190.595465216,"vramReservedPct":99.52609489600644,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 133.81 MiB is free. Including non-PyTorch memory, this process has 178.21 GiB memory in use. Of the allocated memory 177.22 GiB is allocated by PyTorch, and 169.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":120259.11699847983,"meanTps":119912.46206354015,"stepMs":272.4783020019531,"jitter":0.00206576051548686,"achievedTflops":272.60904649076815,"nominalPeakTflops":2250.0,"mfuNominalPct":12.115957621811917,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.115957621811917,"vramAllocatedGb":68.831059968,"vramAllocatedPct":35.942547732729565,"vramReservedGb":71.267516416,"vramReservedPct":37.21482876721705,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":130114.8844714928,"meanTps":130001.27281144331,"stepMs":503.67796325683594,"jitter":0.0008963680584581695,"achievedTflops":294.95056570620324,"nominalPeakTflops":2250.0,"mfuNominalPct":13.10891403138681,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.10891403138681,"vramAllocatedGb":132.764933632,"vramAllocatedPct":69.32785818668671,"vramReservedGb":137.63608576,"vramReservedPct":71.87150080900611,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.409678848,"vramAllocatedPct":99.42908003958792,"vramReservedGb":190.595465216,"vramReservedPct":99.52609489600644,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 133.81 MiB is free. Including non-PyTorch memory, this process has 178.21 GiB memory in use. Of the allocated memory 177.22 GiB is allocated by PyTorch, and 169.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":130602.17414288972,"meanTps":130371.50777754682,"stepMs":501.79869079589844,"jitter":0.0006060531877527129,"achievedTflops":296.0551769489915,"nominalPeakTflops":2250.0,"mfuNominalPct":13.158007864399622,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.158007864399622,"vramAllocatedGb":132.764933632,"vramAllocatedPct":46.190572128531244,"vramReservedGb":137.596239872,"vramReservedPct":47.871443675323135,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":256,"tokensPerStep":131072,"status":"complete","stable":true,"tps":137118.34226716417,"meanTps":137082.15341112416,"stepMs":955.9042053222656,"jitter":0.00026757077347689014,"achievedTflops":310.82633462474973,"nominalPeakTflops":2250.0,"mfuNominalPct":13.814503761099989,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.814503761099989,"vramAllocatedGb":260.63268096,"vramAllocatedPct":90.67735221639637,"vramReservedGb":270.173995008,"vramReservedPct":93.99689407642323,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":512,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.977933824,"vramAllocatedPct":99.49528100605292,"vramReservedGb":286.506614784,"vramReservedPct":99.67921568931462,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 445.62 MiB is free. Including non-PyTorch memory, this process has 267.24 GiB memory in use. Of the allocated memory 266.34 GiB is allocated by PyTorch, and 84.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":130729.66047863735,"meanTps":130722.05480929872,"stepMs":501.30934143066406,"jitter":0.00021746791673379483,"achievedTflops":296.3441689962989,"nominalPeakTflops":2250.0,"mfuNominalPct":13.170851955391063,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.170851955391063,"vramAllocatedGb":132.764933632,"vramAllocatedPct":46.190572128531244,"vramReservedGb":137.596239872,"vramReservedPct":47.871443675323135,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":256,"tokensPerStep":131072,"status":"complete","stable":true,"tps":137222.53268013845,"meanTps":137221.24461236378,"stepMs":955.1784057617188,"jitter":0.00012647387547434788,"achievedTflops":311.0625183740014,"nominalPeakTflops":2250.0,"mfuNominalPct":13.825000816622286,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.825000816622286,"vramAllocatedGb":260.63268096,"vramAllocatedPct":90.67735221639637,"vramReservedGb":270.173995008,"vramReservedPct":93.99689407642323,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":512,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.977933824,"vramAllocatedPct":99.49528100605292,"vramReservedGb":286.506614784,"vramReservedPct":99.67921568931462,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 445.62 MiB is free. Including non-PyTorch memory, this process has 267.24 GiB memory in use. Of the allocated memory 266.34 GiB is allocated by PyTorch, and 84.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":30947.99567292566,"meanTps":30943.387391710647,"stepMs":132.35105895996094,"jitter":0.0007751646085624467,"achievedTflops":70.1543783271195,"nominalPeakTflops":165.2,"mfuNominalPct":42.466330706488804,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":34.13561511334986,"vramAllocatedGb":12.888920576,"vramAllocatedPct":51.022905597439895,"vramReservedGb":13.19108608,"vramReservedPct":52.21907729347028,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":31329.54845191316,"meanTps":31335.718587969754,"stepMs":261.4783935546875,"jitter":0.0008207996406214476,"achievedTflops":71.01930018802923,"nominalPeakTflops":165.2,"mfuNominalPct":42.98989115498138,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":34.5564675312769,"vramAllocatedGb":20.880654848,"vramAllocatedPct":82.65949618046818,"vramReservedGb":21.472739328,"vramReservedPct":85.00335968328174,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.622859264,"vramAllocatedPct":97.47362599979765,"vramReservedGb":24.696061952,"vramReservedPct":97.76341084386424,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 88.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 59.69 MiB is free. Including non-PyTorch memory, this process has 23.46 GiB memory in use. Of the allocated memory 22.93 GiB is allocated by PyTorch, and 69.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":30886.92339967411,"meanTps":30863.4543587388,"stepMs":132.61275482177734,"jitter":0.001597423882966449,"achievedTflops":70.01593681354728,"nominalPeakTflops":165.2,"mfuNominalPct":42.382528337498364,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":34.06825244353935,"vramAllocatedGb":12.888920576,"vramAllocatedPct":51.022905597439895,"vramReservedGb":13.19108608,"vramReservedPct":52.21907729347028,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":31312.25715865618,"meanTps":31308.392389648194,"stepMs":261.62278747558594,"jitter":0.0006904677793764283,"achievedTflops":70.98010346777195,"nominalPeakTflops":165.2,"mfuNominalPct":42.96616432673847,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":34.53739524828738,"vramAllocatedGb":20.880654848,"vramAllocatedPct":82.65949618046818,"vramReservedGb":21.472739328,"vramReservedPct":85.00335968328174,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.622859264,"vramAllocatedPct":97.47362599979765,"vramReservedGb":24.696061952,"vramReservedPct":97.76341084386424,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 88.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 59.69 MiB is free. Including non-PyTorch memory, this process has 23.46 GiB memory in use. Of the allocated memory 22.93 GiB is allocated by PyTorch, and 69.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":40401.623376867705,"meanTps":40394.72517429916,"stepMs":101.38206481933594,"jitter":0.00035658223932993937,"achievedTflops":91.58430812015905,"nominalPeakTflops":209.5,"mfuNominalPct":43.71566020055325,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":34.05294307531769,"vramAllocatedGb":12.888920576,"vramAllocatedPct":38.28143096247951,"vramReservedGb":13.19108608,"vramReservedPct":39.17889385026842,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":44169.92363800206,"meanTps":44167.65487448008,"stepMs":185.46556854248047,"jitter":0.000253652819823226,"achievedTflops":100.12646913645635,"nominalPeakTflops":209.5,"mfuNominalPct":47.79306402694814,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":37.22909550578128,"vramAllocatedGb":20.880654848,"vramAllocatedPct":62.01771066100758,"vramReservedGb":21.472739328,"vramReservedPct":63.776262978203235,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.678009856,"vramAllocatedPct":97.0570786682524,"vramReservedGb":32.807845888,"vramReservedPct":97.44270515001577,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 251.88 MiB is free. Including non-PyTorch memory, this process has 31.10 GiB memory in use. Of the allocated memory 30.43 GiB is allocated by PyTorch, and 83.82 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":40389.98511831657,"meanTps":40387.12703530055,"stepMs":101.4112777709961,"jitter":0.0003826912502457272,"achievedTflops":91.5579259659771,"nominalPeakTflops":209.5,"mfuNominalPct":43.70306728686258,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":34.04313364384409,"vramAllocatedGb":12.888920576,"vramAllocatedPct":38.28143096247951,"vramReservedGb":13.19108608,"vramReservedPct":39.17889385026842,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":44161.14930168507,"meanTps":44152.56109584946,"stepMs":185.5024185180664,"jitter":0.0003818975788319018,"achievedTflops":100.10657905646354,"nominalPeakTflops":209.5,"mfuNominalPct":47.783569955352526,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":37.22169996198496,"vramAllocatedGb":20.880654848,"vramAllocatedPct":62.01771066100758,"vramReservedGb":21.472739328,"vramReservedPct":63.776262978203235,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.678009856,"vramAllocatedPct":97.0570786682524,"vramReservedGb":32.807845888,"vramReservedPct":97.44270515001577,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 251.88 MiB is free. Including non-PyTorch memory, this process has 31.10 GiB memory in use. Of the allocated memory 30.43 GiB is allocated by PyTorch, and 83.82 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":78535.00824761007,"meanTps":78527.90594582396,"stepMs":208.6203384399414,"jitter":0.0009304649435887607,"achievedTflops":178.02686606119204,"nominalPeakTflops":989.5,"mfuNominalPct":17.991598389205866,"configuredPeakTflops":989.5,"mfuConfiguredPct":17.991598389205866,"vramAllocatedGb":36.914192896,"vramAllocatedPct":43.4195262346369,"vramReservedGb":38.088474624,"vramReservedPct":44.80074988591374,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":84262.016963574,"meanTps":84226.89029309682,"stepMs":388.88221740722656,"jitter":0.00037492888682848105,"achievedTflops":191.0091199166148,"nominalPeakTflops":989.5,"mfuNominalPct":19.30359978945071,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.30359978945071,"vramAllocatedGb":68.881129472,"vramAllocatedPct":81.01994852242875,"vramReservedGb":71.328333824,"vramReservedPct":83.8984200594482,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.7314816,"vramAllocatedPct":98.48735612990109,"vramReservedGb":84.16919552,"vramReservedPct":99.00220772333157,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 318.19 MiB is free. Including non-PyTorch memory, this process has 78.86 GiB memory in use. Of the allocated memory 77.98 GiB is allocated by PyTorch, and 157.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":78550.00912064909,"meanTps":78533.88585429855,"stepMs":208.58049774169922,"jitter":0.0005414875045278618,"achievedTflops":178.06087074871814,"nominalPeakTflops":989.5,"mfuNominalPct":17.9950349417603,"configuredPeakTflops":989.5,"mfuConfiguredPct":17.9950349417603,"vramAllocatedGb":36.914192896,"vramAllocatedPct":43.4195262346369,"vramReservedGb":38.088474624,"vramReservedPct":44.80074988591374,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":84276.7223604977,"meanTps":84263.31964065217,"stepMs":388.8143615722656,"jitter":0.00035195355126171164,"achievedTflops":191.04245480492676,"nominalPeakTflops":989.5,"mfuNominalPct":19.30696865133166,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.30696865133166,"vramAllocatedGb":68.881129472,"vramAllocatedPct":81.01994852242875,"vramReservedGb":71.328333824,"vramReservedPct":83.8984200594482,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.7314816,"vramAllocatedPct":98.48735612990109,"vramReservedGb":84.16919552,"vramReservedPct":99.00220772333157,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 318.19 MiB is free. Including non-PyTorch memory, this process has 78.86 GiB memory in use. Of the allocated memory 77.98 GiB is allocated by PyTorch, and 157.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":94234.49644234093,"meanTps":94294.62536345914,"stepMs":347.72828674316406,"jitter":0.0021266322405719914,"achievedTflops":213.61520741923437,"nominalPeakTflops":989.5,"mfuNominalPct":21.58819680841176,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.58819680841176,"vramAllocatedGb":68.881129472,"vramAllocatedPct":45.883813667757906,"vramReservedGb":71.351402496,"vramReservedPct":47.52933760167917,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":97230.95663448202,"meanTps":97259.25446289433,"stepMs":674.0240173339844,"jitter":0.0011582802535802658,"achievedTflops":220.40772491158762,"nominalPeakTflops":989.5,"mfuNominalPct":22.274656383182172,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.274656383182172,"vramAllocatedGb":132.815003136,"vramAllocatedPct":88.47210989262489,"vramReservedGb":137.71997184,"vramReservedPct":91.73945861041857,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.095695872,"vramAllocatedPct":98.6510437107455,"vramReservedGb":149.017329664,"vramReservedPct":99.26497198843478,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.06 GiB is free. Including non-PyTorch memory, this process has 138.74 GiB memory in use. Of the allocated memory 137.92 GiB is allocated by PyTorch, and 98.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":94289.14915282695,"meanTps":94274.8168052956,"stepMs":347.5267333984375,"jitter":0.0015250094136193292,"achievedTflops":213.73909676472093,"nominalPeakTflops":989.5,"mfuNominalPct":21.600717207147138,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.600717207147138,"vramAllocatedGb":68.881129472,"vramAllocatedPct":45.883813667757906,"vramReservedGb":71.351402496,"vramReservedPct":47.52933760167917,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":97361.04220224149,"meanTps":97321.54203289586,"stepMs":673.1234436035156,"jitter":0.001045964264530447,"achievedTflops":220.70260902078627,"nominalPeakTflops":989.5,"mfuNominalPct":22.30445770801276,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.30445770801276,"vramAllocatedGb":132.815003136,"vramAllocatedPct":88.47210989262489,"vramReservedGb":137.71997184,"vramReservedPct":91.73945861041857,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.095695872,"vramAllocatedPct":98.6510437107455,"vramReservedGb":149.017329664,"vramReservedPct":99.26497198843478,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.06 GiB is free. Including non-PyTorch memory, this process has 138.74 GiB memory in use. Of the allocated memory 137.92 GiB is allocated by PyTorch, and 98.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":28543.8620752209,"meanTps":28507.494561233063,"stepMs":143.49845123291016,"jitter":0.004905706364881778,"achievedTflops":64.70457473580427,"nominalPeakTflops":364.2,"mfuNominalPct":17.766220410709575,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.333134341007241,"vramAllocatedGb":12.888920576,"vramAllocatedPct":25.333570389556353,"vramReservedGb":13.19108608,"vramReservedPct":25.927485994803682,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":29017.774931154043,"meanTps":29001.28750221168,"stepMs":282.3097229003906,"jitter":0.00739583123484916,"achievedTflops":65.77886278148553,"nominalPeakTflops":364.2,"mfuNominalPct":18.061192416662696,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.571106925520144,"vramAllocatedGb":20.880654848,"vramAllocatedPct":41.04157025816707,"vramReservedGb":21.472739328,"vramReservedPct":42.205330540666914,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":27387.656165030665,"meanTps":27327.89593608658,"stepMs":598.2257080078125,"jitter":0.002066660596061963,"achievedTflops":62.0836325686675,"nominalPeakTflops":364.2,"mfuNominalPct":17.04657676240184,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":13.752552267251726,"vramAllocatedGb":36.864123392,"vramAllocatedPct":72.4575699953885,"vramReservedGb":38.029754368,"vramReservedPct":74.74865358183942,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":18156.28944742842,"meanTps":18155.462580918647,"stepMs":451.1935119628906,"jitter":0.0013035805236342492,"achievedTflops":41.15753447729392,"nominalPeakTflops":154.8,"mfuNominalPct":26.587554571895296,"configuredPeakTflops":180.6,"mfuConfiguredPct":22.78933249019597,"vramAllocatedGb":20.880654848,"vramAllocatedPct":40.915238394035825,"vramReservedGb":21.487419392,"vramReservedPct":42.10418175560795,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19582.46392914594,"meanTps":19581.381863944367,"stepMs":836.6669311523438,"jitter":0.0004207276492307274,"achievedTflops":44.39045415352432,"nominalPeakTflops":154.8,"mfuNominalPct":28.676003975144905,"configuredPeakTflops":180.6,"mfuConfiguredPct":24.579431978695634,"vramAllocatedGb":36.864123392,"vramAllocatedPct":72.23453515948047,"vramReservedGb":38.027657216,"vramReservedPct":74.51445713199678,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.53278208,"vramAllocatedPct":99.01800688569388,"vramReservedGb":50.658803712,"vramReservedPct":99.26494383061348,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 80.25 MiB is free. Process 1415479 has 47.44 GiB memory in use. Of the allocated memory 47.00 GiB is allocated by PyTorch, and 124.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":18204.86729377225,"meanTps":18199.691858353017,"stepMs":449.9895477294922,"jitter":0.0010218650844178202,"achievedTflops":41.26765303381496,"nominalPeakTflops":154.8,"mfuNominalPct":26.65869059031974,"configuredPeakTflops":180.6,"mfuConfiguredPct":22.850306220274064,"vramAllocatedGb":20.880654848,"vramAllocatedPct":40.915238394035825,"vramReservedGb":21.487419392,"vramReservedPct":42.10418175560795,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19608.75804019988,"meanTps":19606.74381001118,"stepMs":835.5450134277344,"jitter":0.0004470683172323278,"achievedTflops":44.45005888638484,"nominalPeakTflops":154.8,"mfuNominalPct":28.71450832453801,"configuredPeakTflops":180.6,"mfuConfiguredPct":24.61243570674687,"vramAllocatedGb":36.864123392,"vramAllocatedPct":72.23453515948047,"vramReservedGb":38.027657216,"vramReservedPct":74.51445713199678,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.53278208,"vramAllocatedPct":99.01800688569388,"vramReservedGb":50.658803712,"vramReservedPct":99.26494383061348,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 80.25 MiB is free. Process 1449495 has 47.44 GiB memory in use. Of the allocated memory 47.00 GiB is allocated by PyTorch, and 124.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":56724.68383837397,"meanTps":56657.07100958783,"stepMs":144.4168472290039,"jitter":0.000516626711229671,"achievedTflops":128.58619254509316,"nominalPeakTflops":468.0,"mfuNominalPct":27.47568216775495,"configuredPeakTflops":468.0,"mfuConfiguredPct":27.47568216775495,"vramAllocatedGb":20.880654848,"vramAllocatedPct":20.476459677191308,"vramReservedGb":21.487419392,"vramReservedPct":21.071478837711336,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":57005.88852085117,"meanTps":56993.047283612665,"stepMs":287.4089050292969,"jitter":0.00026518249721405653,"achievedTflops":129.22364060119205,"nominalPeakTflops":468.0,"mfuNominalPct":27.611889017348727,"configuredPeakTflops":468.0,"mfuConfiguredPct":27.611889017348727,"vramAllocatedGb":36.864123392,"vramAllocatedPct":36.15052984047548,"vramReservedGb":38.048628736,"vramReservedPct":37.31210624171352,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":51409.23537618153,"meanTps":51407.71230620425,"stepMs":637.3952026367188,"jitter":0.0001840635471345004,"achievedTflops":116.5368829117336,"nominalPeakTflops":468.0,"mfuNominalPct":24.90104335720803,"configuredPeakTflops":468.0,"mfuConfiguredPct":24.90104335720803,"vramAllocatedGb":68.831059968,"vramAllocatedPct":67.4986696649548,"vramReservedGb":71.372374016,"vramReservedPct":69.99079048251316,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":56479.063084766574,"meanTps":56479.81797806035,"stepMs":145.0448989868164,"jitter":0.0003437583019594899,"achievedTflops":128.0294078196565,"nominalPeakTflops":468.0,"mfuNominalPct":27.35671107257618,"configuredPeakTflops":468.0,"mfuConfiguredPct":27.35671107257618,"vramAllocatedGb":21.0017408,"vramAllocatedPct":20.595201720084653,"vramReservedGb":22.72264192,"vramReservedPct":22.282790670173952,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":56783.85554605993,"meanTps":56774.888615280455,"stepMs":288.53271484375,"jitter":0.00044973651032720433,"achievedTflops":128.7203257668738,"nominalPeakTflops":468.0,"mfuNominalPct":27.50434311257987,"configuredPeakTflops":468.0,"mfuConfiguredPct":27.50434311257987,"vramAllocatedGb":36.967408128,"vramAllocatedPct":36.25181525261263,"vramReservedGb":38.87071232,"vramReservedPct":38.11827642562752,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":51379.32215240589,"meanTps":51379.18226199215,"stepMs":637.7662963867188,"jitter":0.00013110355346165046,"achievedTflops":116.46907420321757,"nominalPeakTflops":468.0,"mfuNominalPct":24.88655431692683,"configuredPeakTflops":468.0,"mfuConfiguredPct":24.88655431692683,"vramAllocatedGb":68.962656256,"vramAllocatedPct":67.62771859107879,"vramReservedGb":71.760347136,"vramReservedPct":70.37125345196237,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":36414.26375631141,"meanTps":36381.1022101975,"stepMs":449.93357849121094,"jitter":0.002597768815363677,"achievedTflops":88.96033767077883,"nominalPeakTflops":312.0,"mfuNominalPct":28.51292874063424,"configuredPeakTflops":312.0,"mfuConfiguredPct":28.51292874063424,"vramAllocatedGb":36.8710784,"vramAllocatedPct":43.390889416699316,"vramReservedGb":38.027657216,"vramReservedPct":44.751982872179944,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":38129.10615426392,"meanTps":38124.06418024438,"stepMs":859.3959655761719,"jitter":0.0013445211333089993,"achievedTflops":93.14971136771598,"nominalPeakTflops":312.0,"mfuNominalPct":29.855676720421787,"configuredPeakTflops":312.0,"mfuConfiguredPct":29.855676720421787,"vramAllocatedGb":68.843135488,"vramAllocatedPct":81.01647710577015,"vramReservedGb":71.368179712,"vramReservedPct":83.98801792991537,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.683247104,"vramAllocatedPct":98.48072469504953,"vramReservedGb":84.12725248,"vramReservedPct":99.00324231608109,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 522.75 MiB is free. Process 764857 has 78.62 GiB memory in use. Of the allocated memory 77.94 GiB is allocated by PyTorch, and 183.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":36362.13788580268,"meanTps":36345.729138502036,"stepMs":450.5785675048828,"jitter":0.0024915755184958628,"achievedTflops":88.83299375211905,"nominalPeakTflops":312.0,"mfuNominalPct":28.47211338208944,"configuredPeakTflops":312.0,"mfuConfiguredPct":28.47211338208944,"vramAllocatedGb":36.8710784,"vramAllocatedPct":43.390889416699316,"vramReservedGb":38.027657216,"vramReservedPct":44.751982872179944,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":38075.82158585559,"meanTps":38078.16189952702,"stepMs":860.5986328125,"jitter":0.0015925803656838614,"achievedTflops":93.01953674081791,"nominalPeakTflops":312.0,"mfuNominalPct":29.813954083595483,"configuredPeakTflops":312.0,"mfuConfiguredPct":29.813954083595483,"vramAllocatedGb":68.843135488,"vramAllocatedPct":81.01647710577015,"vramReservedGb":71.368179712,"vramReservedPct":83.98801792991537,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.683247104,"vramAllocatedPct":98.48072469504953,"vramReservedGb":84.12725248,"vramReservedPct":99.00324231608109,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 522.75 MiB is free. Process 852916 has 78.62 GiB memory in use. Of the allocated memory 77.94 GiB is allocated by PyTorch, and 183.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":116045.22680327414,"meanTps":115728.44152949897,"stepMs":282.37266540527344,"jitter":0.001158082523779996,"achievedTflops":283.4994174422132,"nominalPeakTflops":2250.0,"mfuNominalPct":12.599974108542808,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.599974108542808,"vramAllocatedGb":68.843135488,"vramAllocatedPct":35.94885338826065,"vramReservedGb":71.280099328,"vramReservedPct":37.22139937524469,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":125227.39502934135,"meanTps":125150.26293119475,"stepMs":523.3359680175781,"jitter":0.00035576507873233106,"achievedTflops":305.93152787584114,"nominalPeakTflops":2250.0,"mfuNominalPct":13.596956794481828,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.596956794481828,"vramAllocatedGb":132.787249152,"vramAllocatedPct":69.33951101671946,"vramReservedGb":137.659154432,"vramReservedPct":71.88354692372346,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.411517952,"vramAllocatedPct":99.43004039212842,"vramReservedGb":190.603853824,"vramReservedPct":99.5304753013582,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 127.81 MiB is free. Including non-PyTorch memory, this process has 178.21 GiB memory in use. Of the allocated memory 177.23 GiB is allocated by PyTorch, and 173.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":115606.52289538001,"meanTps":115547.99188274959,"stepMs":283.4442138671875,"jitter":0.0016044113954715273,"achievedTflops":282.4276602855966,"nominalPeakTflops":2250.0,"mfuNominalPct":12.552340457137626,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.552340457137626,"vramAllocatedGb":68.843135488,"vramAllocatedPct":35.94885338826065,"vramReservedGb":71.280099328,"vramReservedPct":37.22139937524469,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":125010.06479849272,"meanTps":124922.58655607265,"stepMs":524.2457885742188,"jitter":0.0007460258372012528,"achievedTflops":305.4005883832361,"nominalPeakTflops":2250.0,"mfuNominalPct":13.573359483699383,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.573359483699383,"vramAllocatedGb":132.787249152,"vramAllocatedPct":69.33951101671946,"vramReservedGb":137.659154432,"vramReservedPct":71.88354692372346,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.411517952,"vramAllocatedPct":99.43004039212842,"vramReservedGb":190.603853824,"vramReservedPct":99.5304753013582,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 127.81 MiB is free. Including non-PyTorch memory, this process has 178.21 GiB memory in use. Of the allocated memory 177.23 GiB is allocated by PyTorch, and 173.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":125376.30828035815,"meanTps":125380.95068945394,"stepMs":522.7143859863281,"jitter":0.0005942239468528847,"achievedTflops":306.2953241393812,"nominalPeakTflops":2250.0,"mfuNominalPct":13.61312551730583,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.61312551730583,"vramAllocatedGb":132.787249152,"vramAllocatedPct":46.19833597556485,"vramReservedGb":137.608822784,"vramReservedPct":47.87582142695094,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":128,"tokensPerStep":131072,"status":"complete","stable":true,"tps":131516.67550749212,"meanTps":131515.69449208627,"stepMs":996.6188659667969,"jitter":0.0002711219976650177,"achievedTflops":321.2962903982073,"nominalPeakTflops":2250.0,"mfuNominalPct":14.279835128809214,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.279835128809214,"vramAllocatedGb":260.675476992,"vramAllocatedPct":90.69224148835119,"vramReservedGb":270.226423808,"vramReservedPct":94.01513470820575,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.979768832,"vramAllocatedPct":99.49591942816531,"vramReservedGb":286.508711936,"vramReservedPct":99.67994531458592,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 443.62 MiB is free. Including non-PyTorch memory, this process has 267.24 GiB memory in use. Of the allocated memory 266.34 GiB is allocated by PyTorch, and 84.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":125545.25775850753,"meanTps":125536.2872401555,"stepMs":522.0109558105469,"jitter":0.00021469426793111238,"achievedTflops":306.708069066096,"nominalPeakTflops":2250.0,"mfuNominalPct":13.631469736270935,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.631469736270935,"vramAllocatedGb":132.787249152,"vramAllocatedPct":46.19833597556485,"vramReservedGb":137.608822784,"vramReservedPct":47.87582142695094,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":128,"tokensPerStep":131072,"status":"complete","stable":true,"tps":131616.0045176963,"meanTps":131600.7363201525,"stepMs":995.8667297363281,"jitter":0.0001739635267976669,"achievedTflops":321.53895196476816,"nominalPeakTflops":2250.0,"mfuNominalPct":14.29062008732303,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.29062008732303,"vramAllocatedGb":260.675476992,"vramAllocatedPct":90.69224148835119,"vramReservedGb":270.226423808,"vramReservedPct":94.01513470820575,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.979768832,"vramAllocatedPct":99.49591942816531,"vramReservedGb":286.508711936,"vramReservedPct":99.67994531458592,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 443.62 MiB is free. Including non-PyTorch memory, this process has 267.24 GiB memory in use. Of the allocated memory 266.34 GiB is allocated by PyTorch, and 84.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":30118.507066619924,"meanTps":30107.329106414396,"stepMs":135.9961166381836,"jitter":0.001043584115887529,"achievedTflops":73.57975371180902,"nominalPeakTflops":165.2,"mfuNominalPct":44.539802488988514,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":35.80232927344531,"vramAllocatedGb":12.892035584,"vramAllocatedPct":51.03523686740017,"vramReservedGb":13.193183232,"vramReservedPct":52.22737921354874,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":30655.300807947122,"meanTps":30654.183733123842,"stepMs":267.22947692871094,"jitter":0.00046662597724418637,"achievedTflops":74.89114511622128,"nominalPeakTflops":165.2,"mfuNominalPct":45.333622951707795,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":36.44042419084646,"vramAllocatedGb":20.885049856,"vramAllocatedPct":82.67689454032009,"vramReservedGb":21.493710848,"vramReservedPct":85.08637888406628,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.624694272,"vramAllocatedPct":97.4808901798663,"vramReservedGb":24.696061952,"vramReservedPct":97.76341084386424,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 88.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 59.69 MiB is free. Including non-PyTorch memory, this process has 23.46 GiB memory in use. Of the allocated memory 22.93 GiB is allocated by PyTorch, and 68.06 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":29777.460740736307,"meanTps":29682.154353608097,"stepMs":137.55370330810547,"jitter":0.010527280932035655,"achievedTflops":72.74657480930472,"nominalPeakTflops":165.2,"mfuNominalPct":44.035456906358796,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":35.39692229793434,"vramAllocatedGb":12.892035584,"vramAllocatedPct":51.03523686740017,"vramReservedGb":13.193183232,"vramReservedPct":52.22737921354874,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":30621.720891362234,"meanTps":30625.89561995787,"stepMs":267.52252197265625,"jitter":0.0011396642983977738,"achievedTflops":74.80910911136506,"nominalPeakTflops":165.2,"mfuNominalPct":45.28396435312655,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":36.40050723122135,"vramAllocatedGb":20.885049856,"vramAllocatedPct":82.67689454032009,"vramReservedGb":21.493710848,"vramReservedPct":85.08637888406628,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.624694272,"vramAllocatedPct":97.4808901798663,"vramReservedGb":24.696061952,"vramReservedPct":97.76341084386424,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 88.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 59.69 MiB is free. Including non-PyTorch memory, this process has 23.46 GiB memory in use. Of the allocated memory 22.93 GiB is allocated by PyTorch, and 68.06 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":39345.247012166066,"meanTps":39343.0310499837,"stepMs":104.10406112670898,"jitter":0.0004905735765385849,"achievedTflops":96.12075321269779,"nominalPeakTflops":209.5,"mfuNominalPct":45.8810277864906,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":35.73968734047989,"vramAllocatedGb":12.892035584,"vramAllocatedPct":38.29068285933126,"vramReservedGb":13.193183232,"vramReservedPct":39.18512260922713,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":43133.32950101578,"meanTps":43130.16909472723,"stepMs":189.92273712158203,"jitter":0.0002773001346975259,"achievedTflops":105.37506903761758,"nominalPeakTflops":209.5,"mfuNominalPct":50.29836230912534,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":39.180633682221575,"vramAllocatedGb":20.885049856,"vramAllocatedPct":62.030764290622216,"vramReservedGb":21.472739328,"vramReservedPct":63.776262978203235,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.680868864,"vramAllocatedPct":97.06557021855158,"vramReservedGb":32.807845888,"vramReservedPct":97.44270515001577,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 251.88 MiB is free. Including non-PyTorch memory, this process has 31.10 GiB memory in use. Of the allocated memory 30.44 GiB is allocated by PyTorch, and 81.09 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":39339.02403745331,"meanTps":39333.92430951668,"stepMs":104.12052917480469,"jitter":0.00048093593114411173,"achievedTflops":96.10555043569073,"nominalPeakTflops":209.5,"mfuNominalPct":45.87377109102183,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":35.73403463303868,"vramAllocatedGb":12.892035584,"vramAllocatedPct":38.29068285933126,"vramReservedGb":13.193183232,"vramReservedPct":39.18512260922713,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":43109.37768494676,"meanTps":43105.25361397399,"stepMs":190.02825927734375,"jitter":0.0001982166412980183,"achievedTflops":105.31655455934643,"nominalPeakTflops":209.5,"mfuNominalPct":50.270431770571086,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":39.15887678697884,"vramAllocatedGb":20.885049856,"vramAllocatedPct":62.030764290622216,"vramReservedGb":21.472739328,"vramReservedPct":63.776262978203235,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.680868864,"vramAllocatedPct":97.06557021855158,"vramReservedGb":32.807845888,"vramReservedPct":97.44270515001577,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 251.88 MiB is free. Including non-PyTorch memory, this process has 31.10 GiB memory in use. Of the allocated memory 30.44 GiB is allocated by PyTorch, and 81.09 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":76393.4655929215,"meanTps":76392.11796808624,"stepMs":214.4686050415039,"jitter":0.0005217163535091761,"achievedTflops":186.62984759123196,"nominalPeakTflops":989.5,"mfuNominalPct":18.861025527158358,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.861025527158358,"vramAllocatedGb":36.921147904,"vramAllocatedPct":43.42770691335765,"vramReservedGb":38.090571776,"vramReservedPct":44.80321661589314,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":81940.73337114147,"meanTps":81927.19344410516,"stepMs":399.89878845214844,"jitter":0.00036697617923724666,"achievedTflops":200.18186715156557,"nominalPeakTflops":989.5,"mfuNominalPct":20.230608100208748,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.230608100208748,"vramAllocatedGb":68.893204992,"vramAllocatedPct":81.03415209336727,"vramReservedGb":71.347208192,"vramReservedPct":83.92062062926281,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.733316608,"vramAllocatedPct":98.48951451863306,"vramReservedGb":84.16919552,"vramReservedPct":99.00220772333157,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 318.19 MiB is free. Including non-PyTorch memory, this process has 78.86 GiB memory in use. Of the allocated memory 77.98 GiB is allocated by PyTorch, and 155.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":76449.33097337169,"meanTps":76347.26613112252,"stepMs":214.31188201904297,"jitter":0.0007230833917788023,"achievedTflops":186.76632716259482,"nominalPeakTflops":989.5,"mfuNominalPct":18.874818308498718,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.874818308498718,"vramAllocatedGb":36.921147904,"vramAllocatedPct":43.42770691335765,"vramReservedGb":38.090571776,"vramReservedPct":44.80321661589314,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":81945.192119724,"meanTps":81936.79969448081,"stepMs":399.8770294189453,"jitter":0.0005883140038512053,"achievedTflops":200.1927599100227,"nominalPeakTflops":989.5,"mfuNominalPct":20.231708934817856,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.231708934817856,"vramAllocatedGb":68.893204992,"vramAllocatedPct":81.03415209336727,"vramReservedGb":71.347208192,"vramReservedPct":83.92062062926281,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.733316608,"vramAllocatedPct":98.48951451863306,"vramReservedGb":84.16919552,"vramReservedPct":99.00220772333157,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 318.19 MiB is free. Including non-PyTorch memory, this process has 78.86 GiB memory in use. Of the allocated memory 77.98 GiB is allocated by PyTorch, and 155.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":91378.04839597008,"meanTps":91296.9237663734,"stepMs":358.59815979003906,"jitter":0.0022364142359530656,"achievedTflops":223.23730325574206,"nominalPeakTflops":989.5,"mfuNominalPct":22.560616801995156,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.560616801995156,"vramAllocatedGb":68.893204992,"vramAllocatedPct":45.89185753861003,"vramReservedGb":71.343013888,"vramReservedPct":47.52374969495705,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":94142.31937052919,"meanTps":94163.56857762828,"stepMs":696.1375122070312,"jitter":0.0021402807956886127,"achievedTflops":229.9904393607578,"nominalPeakTflops":989.5,"mfuNominalPct":23.24309644878805,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.24309644878805,"vramAllocatedGb":132.837318656,"vramAllocatedPct":88.4869749386749,"vramReservedGb":137.743040512,"vramReservedPct":91.7548253539044,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.09753088,"vramAllocatedPct":98.65226606534097,"vramReservedGb":149.025718272,"vramReservedPct":99.2705598951569,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.05 GiB is free. Including non-PyTorch memory, this process has 138.75 GiB memory in use. Of the allocated memory 137.93 GiB is allocated by PyTorch, and 105.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":91565.26926555397,"meanTps":91558.75479605784,"stepMs":357.8649444580078,"jitter":0.0018077159729889137,"achievedTflops":223.69468533790246,"nominalPeakTflops":989.5,"mfuNominalPct":22.60684035754446,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.60684035754446,"vramAllocatedGb":68.893204992,"vramAllocatedPct":45.89185753861003,"vramReservedGb":71.343013888,"vramReservedPct":47.52374969495705,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":94222.08852845783,"meanTps":94209.86903864254,"stepMs":695.5481567382812,"jitter":0.001514947900821184,"achievedTflops":230.1853160517307,"nominalPeakTflops":989.5,"mfuNominalPct":23.262790909725183,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.262790909725183,"vramAllocatedGb":132.837318656,"vramAllocatedPct":88.4869749386749,"vramReservedGb":137.743040512,"vramReservedPct":91.7548253539044,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.09753088,"vramAllocatedPct":98.65226606534097,"vramReservedGb":149.025718272,"vramReservedPct":99.2705598951569,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.05 GiB is free. Including non-PyTorch memory, this process has 138.75 GiB memory in use. Of the allocated memory 137.93 GiB is allocated by PyTorch, and 105.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":28172.482447130704,"meanTps":28177.657807382642,"stepMs":145.39009857177734,"jitter":0.006499251035246582,"achievedTflops":68.82559999819988,"nominalPeakTflops":364.2,"mfuNominalPct":18.89774848934648,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.246009650825421,"vramAllocatedGb":12.892035584,"vramAllocatedPct":25.339693033727116,"vramReservedGb":13.193183232,"vramReservedPct":25.931608011655,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":28446.323848609885,"meanTps":28429.331131897437,"stepMs":287.98097229003906,"jitter":0.005990378888579159,"achievedTflops":69.49459673274474,"nominalPeakTflops":364.2,"mfuNominalPct":19.08143787280196,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.394203501246027,"vramAllocatedGb":20.885049856,"vramAllocatedPct":41.05020878176368,"vramReservedGb":21.493710848,"vramReservedPct":42.24655070918012,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":26740.727502751917,"meanTps":26742.4411214133,"stepMs":612.6983642578125,"jitter":0.00447270827667618,"achievedTflops":65.32781121504296,"nominalPeakTflops":364.2,"mfuNominalPct":17.937345199078244,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.47119153742061,"vramAllocatedGb":36.8710784,"vramAllocatedPct":72.47124027783681,"vramReservedGb":38.03185152,"vramReservedPct":74.75277559869075,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":17828.570715067366,"meanTps":17818.34064139505,"stepMs":459.48719787597656,"jitter":0.0010586921780803098,"achievedTflops":43.555340885475296,"nominalPeakTflops":154.8,"mfuNominalPct":28.13652511981608,"configuredPeakTflops":180.6,"mfuConfiguredPct":24.11702153127093,"vramAllocatedGb":20.885049856,"vramAllocatedPct":40.92385032701267,"vramReservedGb":21.489516544,"vramReservedPct":42.10829108429774,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19215.06422398914,"meanTps":19215.020828740107,"stepMs":852.6643371582031,"jitter":0.00048314324516675935,"achievedTflops":46.94255561971925,"nominalPeakTflops":154.8,"mfuNominalPct":30.32464833315197,"configuredPeakTflops":180.6,"mfuConfiguredPct":25.992555714130262,"vramAllocatedGb":36.8710784,"vramAllocatedPct":72.24816336283061,"vramReservedGb":38.050725888,"vramReservedPct":74.55965974758449,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.534617088,"vramAllocatedPct":99.02160254829745,"vramReservedGb":50.66719232,"vramReservedPct":99.28138114537263,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 72.25 MiB is free. Process 1415479 has 47.45 GiB memory in use. Of the allocated memory 47.00 GiB is allocated by PyTorch, and 130.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":17899.333017847024,"meanTps":17900.422300809485,"stepMs":457.6706848144531,"jitter":0.0009513305669491662,"achievedTflops":43.72821375726443,"nominalPeakTflops":154.8,"mfuNominalPct":28.24820010159201,"configuredPeakTflops":180.6,"mfuConfiguredPct":24.212742944221723,"vramAllocatedGb":20.885049856,"vramAllocatedPct":40.92385032701267,"vramReservedGb":21.489516544,"vramReservedPct":42.10829108429774,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19267.668070254822,"meanTps":19268.82653070893,"stepMs":850.33642578125,"jitter":0.0007359305958866234,"achievedTflops":47.07106723697709,"nominalPeakTflops":154.8,"mfuNominalPct":30.40766617375781,"configuredPeakTflops":180.6,"mfuConfiguredPct":26.063713863220986,"vramAllocatedGb":36.8710784,"vramAllocatedPct":72.24816336283061,"vramReservedGb":38.050725888,"vramReservedPct":74.55965974758449,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.534617088,"vramAllocatedPct":99.02160254829745,"vramReservedGb":50.66719232,"vramReservedPct":99.28138114537263,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 72.25 MiB is free. Process 1449495 has 47.45 GiB memory in use. Of the allocated memory 47.00 GiB is allocated by PyTorch, and 130.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":55614.84637392066,"meanTps":55606.0095486564,"stepMs":147.29879760742188,"jitter":0.00047004262965466787,"achievedTflops":135.86751461025918,"nominalPeakTflops":468.0,"mfuNominalPct":29.031520215867346,"configuredPeakTflops":468.0,"mfuConfiguredPct":29.031520215867346,"vramAllocatedGb":20.885049856,"vramAllocatedPct":20.4807696092671,"vramReservedGb":21.489516544,"vramReservedPct":21.07353539430295,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":56087.366892369,"meanTps":56092.04228060452,"stepMs":292.1156921386719,"jitter":0.00024123883846370584,"achievedTflops":137.0218860169926,"nominalPeakTflops":468.0,"mfuNominalPct":29.278180772861667,"configuredPeakTflops":468.0,"mfuConfiguredPct":29.278180772861667,"vramAllocatedGb":36.8710784,"vramAllocatedPct":36.15735021760939,"vramReservedGb":38.050725888,"vramReservedPct":37.31416279830514,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":50829.39645318647,"meanTps":50830.05287047435,"stepMs":644.6663208007812,"jitter":0.0001316654757861823,"achievedTflops":124.17662217030617,"nominalPeakTflops":468.0,"mfuNominalPct":26.53346627570645,"configuredPeakTflops":468.0,"mfuConfiguredPct":26.53346627570645,"vramAllocatedGb":68.843135488,"vramAllocatedPct":67.51051143429397,"vramReservedGb":71.372374016,"vramReservedPct":69.99079048251316,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":55416.87920932788,"meanTps":55396.432036067396,"stepMs":147.8249969482422,"jitter":0.000495456994161455,"achievedTflops":135.3838792434217,"nominalPeakTflops":468.0,"mfuNominalPct":28.928179325517455,"configuredPeakTflops":468.0,"mfuConfiguredPct":28.928179325517455,"vramAllocatedGb":21.008183808,"vramAllocatedPct":20.601520008206947,"vramReservedGb":22.724739072,"vramReservedPct":22.28484722676557,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":56079.22630499508,"meanTps":56074.0057385543,"stepMs":292.15809631347656,"jitter":0.00025258647717567247,"achievedTflops":137.00199849691342,"nominalPeakTflops":468.0,"mfuNominalPct":29.273931302759276,"configuredPeakTflops":468.0,"mfuConfiguredPct":29.273931302759276,"vramAllocatedGb":36.974862848,"vramAllocatedPct":36.2591256686219,"vramReservedGb":38.88119808,"vramReservedPct":38.12855920858561,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":50800.48327890839,"meanTps":50790.4241722615,"stepMs":645.0332336425781,"jitter":0.0002457152476274515,"achievedTflops":124.10598705423942,"nominalPeakTflops":468.0,"mfuNominalPct":26.518373302187907,"configuredPeakTflops":468.0,"mfuConfiguredPct":26.518373302187907,"vramAllocatedGb":68.9736832,"vramAllocatedPct":67.63853208212215,"vramReservedGb":71.777124352,"vramReservedPct":70.38770590469531,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":22412.24381801424,"meanTps":22322.502158682953,"stepMs":91.37862396240234,"jitter":0.008072974905025631,"achievedTflops":62.64960711252862,"nominalPeakTflops":312.0,"mfuNominalPct":20.080002279656608,"configuredPeakTflops":312.0,"mfuConfiguredPct":20.080002279656608,"vramAllocatedGb":8.89950208,"vramAllocatedPct":10.473176467140314,"vramReservedGb":8.969519104,"vramReservedPct":10.555574408223327,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":27456.51925653189,"meanTps":27449.856942143793,"stepMs":149.18132781982422,"jitter":0.004652029747111705,"achievedTflops":76.75001923353636,"nominalPeakTflops":312.0,"mfuNominalPct":24.599365138953964,"configuredPeakTflops":312.0,"mfuConfiguredPct":24.599365138953964,"vramAllocatedGb":12.8963456,"vramAllocatedPct":15.176770794128354,"vramReservedGb":13.230931968,"vramReservedPct":15.570521146009114,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":31684.666369652314,"meanTps":31644.153933001453,"stepMs":258.5477752685547,"jitter":0.004175366419173726,"achievedTflops":88.56908374139502,"nominalPeakTflops":312.0,"mfuNominalPct":28.387526840190713,"configuredPeakTflops":312.0,"mfuConfiguredPct":28.387526840190713,"vramAllocatedGb":20.889999872,"vramAllocatedPct":24.583920885829443,"vramReservedGb":21.495808,"vramReservedPct":25.29685239286629,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":34574.91048290288,"meanTps":34591.25280794116,"stepMs":473.86962890625,"jitter":0.0020195103865592976,"achievedTflops":96.64826847741443,"nominalPeakTflops":312.0,"mfuNominalPct":30.977009127376423,"configuredPeakTflops":312.0,"mfuConfiguredPct":30.977009127376423,"vramAllocatedGb":36.877308416,"vramAllocatedPct":43.398221069231624,"vramReservedGb":38.05282304,"vramReservedPct":44.78159869937159,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":36122.633869566074,"meanTps":36138.79942376057,"stepMs":907.132080078125,"jitter":0.0010115130397056959,"achievedTflops":100.97466537371201,"nominalPeakTflops":312.0,"mfuNominalPct":32.36367479926667,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.36367479926667,"vramAllocatedGb":68.851925504,"vramAllocatedPct":81.02682143603599,"vramReservedGb":71.420608512,"vramReservedPct":84.04971756989798,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.68691712,"vramAllocatedPct":98.48504366984831,"vramReservedGb":84.120961024,"vramReservedPct":98.99583835928317,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 528.75 MiB is free. Process 703773 has 78.61 GiB memory in use. Of the allocated memory 77.94 GiB is allocated by PyTorch, and 173.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":21879.01005828796,"meanTps":21743.48858812029,"stepMs":93.60569763183594,"jitter":0.01000605997871398,"achievedTflops":61.15904303437342,"nominalPeakTflops":312.0,"mfuNominalPct":19.60225738281199,"configuredPeakTflops":312.0,"mfuConfiguredPct":19.60225738281199,"vramAllocatedGb":8.89950208,"vramAllocatedPct":10.473176467140314,"vramReservedGb":8.969519104,"vramReservedPct":10.555574408223327,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":27481.525082562934,"meanTps":27431.777269982227,"stepMs":149.04558563232422,"jitter":0.006231021755398308,"achievedTflops":76.81991875761301,"nominalPeakTflops":312.0,"mfuNominalPct":24.621768832568275,"configuredPeakTflops":312.0,"mfuConfiguredPct":24.621768832568275,"vramAllocatedGb":12.8963456,"vramAllocatedPct":15.176770794128354,"vramReservedGb":13.230931968,"vramReservedPct":15.570521146009114,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":31634.565791028966,"meanTps":31662.780162889972,"stepMs":258.9572448730469,"jitter":0.0028971421874298435,"achievedTflops":88.42903611419847,"nominalPeakTflops":312.0,"mfuNominalPct":28.342639780191814,"configuredPeakTflops":312.0,"mfuConfiguredPct":28.342639780191814,"vramAllocatedGb":20.889999872,"vramAllocatedPct":24.583920885829443,"vramReservedGb":21.495808,"vramReservedPct":25.29685239286629,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":34555.20583329142,"meanTps":34562.449401033744,"stepMs":474.1398468017578,"jitter":0.0020552416679965897,"achievedTflops":96.59318748835898,"nominalPeakTflops":312.0,"mfuNominalPct":30.95935496421762,"configuredPeakTflops":312.0,"mfuConfiguredPct":30.95935496421762,"vramAllocatedGb":36.877308416,"vramAllocatedPct":43.398221069231624,"vramReservedGb":38.05282304,"vramReservedPct":44.78159869937159,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":36144.23049702222,"meanTps":36149.637279773466,"stepMs":906.5900573730469,"jitter":0.0007036414575444027,"achievedTflops":101.03503506431815,"nominalPeakTflops":312.0,"mfuNominalPct":32.38302405907633,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.38302405907633,"vramAllocatedGb":68.851925504,"vramAllocatedPct":81.02682143603599,"vramReservedGb":71.420608512,"vramReservedPct":84.04971756989798,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.68691712,"vramAllocatedPct":98.48504366984831,"vramReservedGb":84.120961024,"vramReservedPct":98.99583835928317,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 528.75 MiB is free. Process 736416 has 78.61 GiB memory in use. Of the allocated memory 77.94 GiB is allocated by PyTorch, and 173.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":12171.183125314752,"meanTps":12150.420093944831,"stepMs":168.26630401611328,"jitter":0.019132429474838834,"achievedTflops":34.022467678257115,"nominalPeakTflops":2250.0,"mfuNominalPct":1.512109674589205,"configuredPeakTflops":2250.0,"mfuConfiguredPct":1.512109674589205,"vramAllocatedGb":8.89950208,"vramAllocatedPct":4.647186581997081,"vramReservedGb":8.969519104,"vramReservedPct":4.683748422369635,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":23947.099601111324,"meanTps":23916.722139035075,"stepMs":171.0436782836914,"jitter":0.020580258099400076,"achievedTflops":66.9400348165203,"nominalPeakTflops":2250.0,"mfuNominalPct":2.975112658512013,"configuredPeakTflops":2250.0,"mfuConfiguredPct":2.975112658512013,"vramAllocatedGb":12.8963456,"vramAllocatedPct":6.734278355168057,"vramReservedGb":13.230931968,"vramReservedPct":6.908994341063837,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":47383.00507490742,"meanTps":47153.0039568643,"stepMs":172.8889923095703,"jitter":0.011808574264418201,"achievedTflops":132.4511135903266,"nominalPeakTflops":2250.0,"mfuNominalPct":5.886716159570072,"configuredPeakTflops":2250.0,"mfuConfiguredPct":5.886716159570072,"vramAllocatedGb":20.889999872,"vramAllocatedPct":10.908444790551602,"vramReservedGb":21.495808,"vramReservedPct":11.22478871388561,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":90765.63277166836,"meanTps":90428.62606634531,"stepMs":180.50885009765625,"jitter":0.0065445271647229354,"achievedTflops":253.71985414037414,"nominalPeakTflops":2250.0,"mfuNominalPct":11.276437961794407,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.276437961794407,"vramAllocatedGb":36.877308416,"vramAllocatedPct":19.256777661318694,"vramReservedGb":38.05282304,"vramReservedPct":19.870613776922383,"warmupSteps":15,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":107192.14334256848,"meanTps":107218.99148734141,"stepMs":305.6940460205078,"jitter":0.0013992737233992371,"achievedTflops":299.6373643126274,"nominalPeakTflops":2250.0,"mfuNominalPct":13.31721619167233,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.31721619167233,"vramAllocatedGb":68.851925504,"vramAllocatedPct":35.953443402852876,"vramReservedGb":71.420608512,"vramReservedPct":37.29477116488667,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":115512.1397056142,"meanTps":115482.8431439637,"stepMs":567.3516235351562,"jitter":0.0005481902243120292,"achievedTflops":322.89449588566174,"nominalPeakTflops":2250.0,"mfuNominalPct":14.350866483807188,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.350866483807188,"vramAllocatedGb":132.80115968,"vramAllocatedPct":69.34677488592123,"vramReservedGb":137.669640192,"vramReservedPct":71.88902243041315,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.41519616,"vramAllocatedPct":99.43196109720941,"vramReservedGb":190.564007936,"vramReservedPct":99.50966837593734,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 183.81 MiB is free. Including non-PyTorch memory, this process has 178.16 GiB memory in use. Of the allocated memory 177.23 GiB is allocated by PyTorch, and 111.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11832.963719609026,"meanTps":11699.906101477482,"stepMs":173.0758285522461,"jitter":0.031128698786157347,"achievedTflops":33.07703298383953,"nominalPeakTflops":2250.0,"mfuNominalPct":1.4700903548373125,"configuredPeakTflops":2250.0,"mfuConfiguredPct":1.4700903548373125,"vramAllocatedGb":8.89950208,"vramAllocatedPct":4.647186581997081,"vramReservedGb":8.969519104,"vramReservedPct":4.683748422369635,"warmupSteps":20,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22692.451715589887,"meanTps":22471.310285616535,"stepMs":180.50054931640625,"jitter":0.03567824016229554,"achievedTflops":63.43288052484228,"nominalPeakTflops":2250.0,"mfuNominalPct":2.819239134437435,"configuredPeakTflops":2250.0,"mfuConfiguredPct":2.819239134437435,"vramAllocatedGb":12.8963456,"vramAllocatedPct":6.734278355168057,"vramReservedGb":13.230931968,"vramReservedPct":6.908994341063837,"warmupSteps":22,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":45332.82118289779,"meanTps":45075.36099415589,"stepMs":180.70792388916016,"jitter":0.03699228228754767,"achievedTflops":126.72017400276064,"nominalPeakTflops":2250.0,"mfuNominalPct":5.632007733456029,"configuredPeakTflops":2250.0,"mfuConfiguredPct":5.632007733456029,"vramAllocatedGb":20.889999872,"vramAllocatedPct":10.908444790551602,"vramReservedGb":21.495808,"vramReservedPct":11.22478871388561,"warmupSteps":16,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":89368.46158606319,"meanTps":88449.8762925874,"stepMs":183.33089447021484,"jitter":0.010821209663253327,"achievedTflops":249.81430025840385,"nominalPeakTflops":2250.0,"mfuNominalPct":11.102857789262394,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.102857789262394,"vramAllocatedGb":36.877308416,"vramAllocatedPct":19.256777661318694,"vramReservedGb":38.05282304,"vramReservedPct":19.870613776922383,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":106650.48420930095,"meanTps":106453.09222909858,"stepMs":307.2466125488281,"jitter":0.004519358260650703,"achievedTflops":298.1232485389606,"nominalPeakTflops":2250.0,"mfuNominalPct":13.249922157287136,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.249922157287136,"vramAllocatedGb":68.851925504,"vramAllocatedPct":35.953443402852876,"vramReservedGb":71.420608512,"vramReservedPct":37.29477116488667,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":115288.95971093951,"meanTps":115248.29574379355,"stepMs":568.4499206542969,"jitter":0.001035601158256199,"achievedTflops":322.2706342547032,"nominalPeakTflops":2250.0,"mfuNominalPct":14.323139300209032,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.323139300209032,"vramAllocatedGb":132.80115968,"vramAllocatedPct":69.34677488592123,"vramReservedGb":137.669640192,"vramReservedPct":71.88902243041315,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.41519616,"vramAllocatedPct":99.43196109720941,"vramReservedGb":190.564007936,"vramReservedPct":99.50966837593734,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 183.81 MiB is free. Including non-PyTorch memory, this process has 178.16 GiB memory in use. Of the allocated memory 177.23 GiB is allocated by PyTorch, and 111.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":24741.361207130467,"meanTps":23849.583380422042,"stepMs":82.7763671875,"jitter":0.003404824601019902,"achievedTflops":69.16025774313641,"nominalPeakTflops":2250.0,"mfuNominalPct":3.0737892330282848,"configuredPeakTflops":2250.0,"mfuConfiguredPct":3.0737892330282848,"vramAllocatedGb":8.89950208,"vramAllocatedPct":3.09624749162812,"vramReservedGb":8.969519104,"vramReservedPct":3.1206072853539357,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":49329.57094960251,"meanTps":49317.02367403725,"stepMs":83.03335952758789,"jitter":0.001231895845290629,"achievedTflops":137.89240667363146,"nominalPeakTflops":2250.0,"mfuNominalPct":6.128551407716953,"configuredPeakTflops":2250.0,"mfuConfiguredPct":6.128551407716953,"vramAllocatedGb":12.8963456,"vramAllocatedPct":4.486799076647819,"vramReservedGb":13.230931968,"vramReservedPct":4.603205836637358,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":76554.84148981026,"meanTps":76580.41818911005,"stepMs":107.00825500488281,"jitter":0.0008239728597028359,"achievedTflops":213.99600954026488,"nominalPeakTflops":2250.0,"mfuNominalPct":9.510933757345107,"configuredPeakTflops":2250.0,"mfuConfiguredPct":9.510933757345107,"vramAllocatedGb":20.889999872,"vramAllocatedPct":7.267890846292352,"vramReservedGb":21.495808,"vramReservedPct":7.478659030834192,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":94493.94851305654,"meanTps":94478.28441122528,"stepMs":173.3867645263672,"jitter":0.0004803446162240415,"achievedTflops":264.1417252518102,"nominalPeakTflops":2250.0,"mfuNominalPct":11.739632233413786,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.739632233413786,"vramAllocatedGb":36.877308416,"vramAllocatedPct":12.830074385581419,"vramReservedGb":38.05282304,"vramReservedPct":13.239050547754772,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":106973.11878305855,"meanTps":106902.13996566064,"stepMs":306.3199462890625,"jitter":0.0002995482138366835,"achievedTflops":299.02511849231996,"nominalPeakTflops":2250.0,"mfuNominalPct":13.290005266325332,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.290005266325332,"vramAllocatedGb":68.851925504,"vramAllocatedPct":23.95444146415955,"vramReservedGb":71.420608512,"vramReservedPct":24.84811823942334,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":116117.70006799493,"meanTps":116113.56697305544,"stepMs":564.3928527832031,"jitter":0.000373335074436203,"achievedTflops":324.58723665245543,"nominalPeakTflops":2250.0,"mfuNominalPct":14.426099406775798,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.426099406775798,"vramAllocatedGb":132.80115968,"vramAllocatedPct":46.20317562131582,"vramReservedGb":137.669640192,"vramReservedPct":47.89698055981867,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":64,"tokensPerStep":131072,"status":"complete","stable":true,"tps":121561.49708524728,"meanTps":121544.08196594588,"stepMs":1078.2361450195312,"jitter":0.00016398900453027345,"achievedTflops":339.80444324276965,"nominalPeakTflops":2250.0,"mfuNominalPct":15.10241969967865,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.10241969967865,"vramAllocatedGb":260.69962752,"vramAllocatedPct":90.70064375749719,"vramReservedGb":270.088011776,"vramReservedPct":93.9669794402999,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":24942.597881887494,"meanTps":24842.564707096775,"stepMs":82.10852813720703,"jitter":0.002023224670508967,"achievedTflops":69.72278056381116,"nominalPeakTflops":2250.0,"mfuNominalPct":3.098790247280496,"configuredPeakTflops":2250.0,"mfuConfiguredPct":3.098790247280496,"vramAllocatedGb":8.89950208,"vramAllocatedPct":3.09624749162812,"vramReservedGb":8.969519104,"vramReservedPct":3.1206072853539357,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":49604.8088873276,"meanTps":49591.19886526529,"stepMs":82.57263946533203,"jitter":0.00393171923099418,"achievedTflops":138.66178741038215,"nominalPeakTflops":2250.0,"mfuNominalPct":6.162746107128095,"configuredPeakTflops":2250.0,"mfuConfiguredPct":6.162746107128095,"vramAllocatedGb":12.8963456,"vramAllocatedPct":4.486799076647819,"vramReservedGb":13.230931968,"vramReservedPct":4.603205836637358,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":76682.61408183468,"meanTps":76622.00531315837,"stepMs":106.82995223999023,"jitter":0.0012361692166612253,"achievedTflops":214.35317604064196,"nominalPeakTflops":2250.0,"mfuNominalPct":9.52680782402853,"configuredPeakTflops":2250.0,"mfuConfiguredPct":9.52680782402853,"vramAllocatedGb":20.889999872,"vramAllocatedPct":7.267890846292352,"vramReservedGb":21.495808,"vramReservedPct":7.478659030834192,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":94558.61126401322,"meanTps":94552.43783401755,"stepMs":173.26819610595703,"jitter":0.0003198830995681357,"achievedTflops":264.32247894943845,"nominalPeakTflops":2250.0,"mfuNominalPct":11.747665731086153,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.747665731086153,"vramAllocatedGb":36.877308416,"vramAllocatedPct":12.830074385581419,"vramReservedGb":38.05282304,"vramReservedPct":13.239050547754772,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":106994.14454827177,"meanTps":106498.88615946598,"stepMs":306.25975036621094,"jitter":0.0002937715481495502,"achievedTflops":299.0838924348374,"nominalPeakTflops":2250.0,"mfuNominalPct":13.29261744154833,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.29261744154833,"vramAllocatedGb":68.851925504,"vramAllocatedPct":23.95444146415955,"vramReservedGb":71.420608512,"vramReservedPct":24.84811823942334,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":116118.8553529794,"meanTps":116048.61189348076,"stepMs":564.3872375488281,"jitter":0.0004465314695849283,"achievedTflops":324.59046605469507,"nominalPeakTflops":2250.0,"mfuNominalPct":14.426242935764225,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.426242935764225,"vramAllocatedGb":132.80115968,"vramAllocatedPct":46.20317562131582,"vramReservedGb":137.669640192,"vramReservedPct":47.89698055981867,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":64,"tokensPerStep":131072,"status":"complete","stable":true,"tps":121469.21502630034,"meanTps":121457.64720080438,"stepMs":1079.0552978515625,"jitter":0.0002504091295316847,"achievedTflops":339.54648447775236,"nominalPeakTflops":2250.0,"mfuNominalPct":15.090954865677883,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.090954865677883,"vramAllocatedGb":260.69962752,"vramAllocatedPct":90.70064375749719,"vramReservedGb":270.088011776,"vramReservedPct":93.9669794402999,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":24064.98984051878,"meanTps":24060.270689708108,"stepMs":85.1028823852539,"jitter":0.0022447132910578186,"achievedTflops":67.26957688474208,"nominalPeakTflops":165.2,"mfuNominalPct":40.72008286001337,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":32.731932633886345,"vramAllocatedGb":8.89950208,"vramAllocatedPct":35.23013830479981,"vramReservedGb":8.969519104,"vramReservedPct":35.5073121755441,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":28678.123901442545,"meanTps":28675.655450356237,"stepMs":142.8266372680664,"jitter":0.0005882700705795704,"achievedTflops":80.1648067787699,"nominalPeakTflops":165.2,"mfuNominalPct":48.52591209368638,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":39.006474793011066,"vramAllocatedGb":12.8963456,"vramAllocatedPct":51.05229877248328,"vramReservedGb":13.230931968,"vramReservedPct":52.37681377496089,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":29328.819942973685,"meanTps":29249.568766135842,"stepMs":279.3157043457031,"jitter":0.0006919251510251111,"achievedTflops":81.98371664262018,"nominalPeakTflops":165.2,"mfuNominalPct":49.62694712023014,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":39.89151730238595,"vramAllocatedGb":20.889999872,"vramAllocatedPct":82.69648999034902,"vramReservedGb":21.495808,"vramReservedPct":85.09468080414473,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.628364288,"vramAllocatedPct":97.49541854000358,"vramReservedGb":24.698159104,"vramReservedPct":97.77171276394269,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 88.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 57.69 MiB is free. Including non-PyTorch memory, this process has 23.46 GiB memory in use. Of the allocated memory 22.94 GiB is allocated by PyTorch, and 66.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":24093.405081627843,"meanTps":24086.53244924225,"stepMs":85.00251388549805,"jitter":0.001941234737903717,"achievedTflops":67.34900684748679,"nominalPeakTflops":165.2,"mfuNominalPct":40.76816395126319,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":32.77058154934075,"vramAllocatedGb":8.89950208,"vramAllocatedPct":35.23013830479981,"vramReservedGb":8.969519104,"vramReservedPct":35.5073121755441,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":28694.124371316637,"meanTps":28683.37684370937,"stepMs":142.7469940185547,"jitter":0.0010806816669511827,"achievedTflops":80.20953336479883,"nominalPeakTflops":165.2,"mfuNominalPct":48.552986298304376,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":39.02823779002457,"vramAllocatedGb":12.8963456,"vramAllocatedPct":51.05229877248328,"vramReservedGb":13.230931968,"vramReservedPct":52.37681377496089,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":29329.64350178533,"meanTps":29326.737378723563,"stepMs":279.307861328125,"jitter":0.00037687207452504054,"achievedTflops":81.986018760891,"nominalPeakTflops":165.2,"mfuNominalPct":49.62834065429238,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":39.892637463737415,"vramAllocatedGb":20.889999872,"vramAllocatedPct":82.69648999034902,"vramReservedGb":21.495808,"vramReservedPct":85.09468080414473,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.628364288,"vramAllocatedPct":97.49541854000358,"vramReservedGb":24.698159104,"vramReservedPct":97.77171276394269,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 88.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 57.69 MiB is free. Including non-PyTorch memory, this process has 23.46 GiB memory in use. Of the allocated memory 22.94 GiB is allocated by PyTorch, and 66.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":29681.42197249678,"meanTps":29669.98163179969,"stepMs":68.9993896484375,"jitter":0.0006229488910079382,"achievedTflops":82.96935551019963,"nominalPeakTflops":209.5,"mfuNominalPct":39.60351098338885,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":30.849725222335586,"vramAllocatedGb":8.89950208,"vramAllocatedPct":26.432444242874883,"vramReservedGb":8.969519104,"vramReservedPct":26.640402066390784,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":37543.82473769298,"meanTps":37543.80084592606,"stepMs":109.09916687011719,"jitter":0.00024157309106685482,"achievedTflops":104.94736218368064,"nominalPeakTflops":209.5,"mfuNominalPct":50.094206292926316,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":39.02160341329244,"vramAllocatedGb":12.8963456,"vramAllocatedPct":38.30348405389044,"vramReservedGb":13.230931968,"vramReservedPct":39.297240270483854,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":41345.51016493124,"meanTps":41341.28290706954,"stepMs":198.1351776123047,"jitter":0.00036257227673774337,"achievedTflops":115.57432574502053,"nominalPeakTflops":209.5,"mfuNominalPct":55.16674259905514,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":42.97292862004069,"vramAllocatedGb":20.889999872,"vramAllocatedPct":62.04546635156284,"vramReservedGb":21.495808,"vramReservedPct":63.84477932674901,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.68505088,"vramAllocatedPct":97.07799123983447,"vramReservedGb":32.80994304,"vramReservedPct":97.44893390897447,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 247.88 MiB is free. Including non-PyTorch memory, this process has 31.11 GiB memory in use. Of the allocated memory 30.44 GiB is allocated by PyTorch, and 79.11 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":29675.103958311764,"meanTps":29672.67360930256,"stepMs":69.01408004760742,"jitter":0.0005354581099864594,"achievedTflops":82.95169457853939,"nominalPeakTflops":209.5,"mfuNominalPct":39.59508094441021,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":30.8431585220697,"vramAllocatedGb":8.89950208,"vramAllocatedPct":26.432444242874883,"vramReservedGb":8.969519104,"vramReservedPct":26.640402066390784,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":37541.379271162004,"meanTps":37516.300725311696,"stepMs":109.10627365112305,"jitter":0.0004927044351259923,"achievedTflops":104.94052629885729,"nominalPeakTflops":209.5,"mfuNominalPct":50.09094334074334,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":39.01906168969877,"vramAllocatedGb":12.8963456,"vramAllocatedPct":38.30348405389044,"vramReservedGb":13.230931968,"vramReservedPct":39.297240270483854,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":41232.0069255707,"meanTps":41232.695681879544,"stepMs":198.68060302734375,"jitter":0.000650508266355532,"achievedTflops":115.25704678760432,"nominalPeakTflops":209.5,"mfuNominalPct":55.01529679599252,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":42.854957730729446,"vramAllocatedGb":20.889999872,"vramAllocatedPct":62.04546635156284,"vramReservedGb":21.495808,"vramReservedPct":63.84477932674901,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.68505088,"vramAllocatedPct":97.07799123983447,"vramReservedGb":32.80994304,"vramReservedPct":97.44893390897447,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 247.88 MiB is free. Including non-PyTorch memory, this process has 31.11 GiB memory in use. Of the allocated memory 30.44 GiB is allocated by PyTorch, and 79.11 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":33354.3009824955,"meanTps":33350.282622153085,"stepMs":61.401376724243164,"jitter":0.006289581347915421,"achievedTflops":93.23626268900348,"nominalPeakTflops":989.5,"mfuNominalPct":9.422563182314653,"configuredPeakTflops":989.5,"mfuConfiguredPct":9.422563182314653,"vramAllocatedGb":8.949571584,"vramAllocatedPct":10.526741280109523,"vramReservedGb":9.032433664,"vramReservedPct":10.624206021287879,"warmupSteps":16,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":54113.81379575852,"meanTps":53998.58364974105,"stepMs":75.69231796264648,"jitter":0.000598949183629891,"achievedTflops":151.26594200888806,"nominalPeakTflops":989.5,"mfuNominalPct":15.287108843748163,"configuredPeakTflops":989.5,"mfuConfiguredPct":15.287108843748163,"vramAllocatedGb":12.946415104,"vramAllocatedPct":15.227942592062988,"vramReservedGb":13.293846528,"vramReservedPct":15.636601339434378,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":65527.133199810545,"meanTps":65532.762759915786,"stepMs":125.01691436767578,"jitter":0.0004502661535606418,"achievedTflops":183.1699308428366,"nominalPeakTflops":989.5,"mfuNominalPct":18.511362389372067,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.511362389372067,"vramAllocatedGb":20.940069376,"vramAllocatedPct":24.63030667331399,"vramReservedGb":21.53775104,"vramReservedPct":25.333316888466804,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":72448.75128608056,"meanTps":72439.38398225686,"stepMs":226.14606475830078,"jitter":0.0005844585972391266,"achievedTflops":202.51813431629893,"nominalPeakTflops":989.5,"mfuNominalPct":20.466713927872554,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.466713927872554,"vramAllocatedGb":36.92737792,"vramAllocatedPct":43.435034835816,"vramReservedGb":38.09476608,"vramReservedPct":44.80815007585195,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":77510.22512930591,"meanTps":77509.13435410771,"stepMs":422.7571258544922,"jitter":0.0001837610312147152,"achievedTflops":216.66662164595823,"nominalPeakTflops":989.5,"mfuNominalPct":21.896576214851766,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.896576214851766,"vramAllocatedGb":68.901995008,"vramAllocatedPct":81.04449116082,"vramReservedGb":71.420608512,"vramReservedPct":84.00695617854191,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.736986624,"vramAllocatedPct":98.49383129609701,"vramReservedGb":84.099989504,"vramReservedPct":98.92080563401127,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 262.19 MiB is free. Including non-PyTorch memory, this process has 78.91 GiB memory in use. Of the allocated memory 77.99 GiB is allocated by PyTorch, and 206.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":32558.326286429827,"meanTps":32427.43660655144,"stepMs":62.902496337890625,"jitter":0.015451999398162905,"achievedTflops":91.01125111118249,"nominalPeakTflops":989.5,"mfuNominalPct":9.197700971317078,"configuredPeakTflops":989.5,"mfuConfiguredPct":9.197700971317078,"vramAllocatedGb":8.949571584,"vramAllocatedPct":10.526741280109523,"vramReservedGb":9.032433664,"vramReservedPct":10.624206021287879,"warmupSteps":26,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":53874.934239466034,"meanTps":53872.46122117826,"stepMs":76.02793502807617,"jitter":0.0009588045860098112,"achievedTflops":150.5981949296371,"nominalPeakTflops":989.5,"mfuNominalPct":15.219625561357967,"configuredPeakTflops":989.5,"mfuConfiguredPct":15.219625561357967,"vramAllocatedGb":12.946415104,"vramAllocatedPct":15.227942592062988,"vramReservedGb":13.293846528,"vramReservedPct":15.636601339434378,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":65352.328206973485,"meanTps":65363.89913556764,"stepMs":125.35131072998047,"jitter":0.0006623855245210193,"achievedTflops":182.6812932833189,"nominalPeakTflops":989.5,"mfuNominalPct":18.461980119587558,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.461980119587558,"vramAllocatedGb":20.940069376,"vramAllocatedPct":24.63030667331399,"vramReservedGb":21.53775104,"vramReservedPct":25.333316888466804,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":72331.4160113368,"meanTps":72333.86526889606,"stepMs":226.5129165649414,"jitter":0.0006570690022712068,"achievedTflops":202.19014355719307,"nominalPeakTflops":989.5,"mfuNominalPct":20.433566807194854,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.433566807194854,"vramAllocatedGb":36.92737792,"vramAllocatedPct":43.435034835816,"vramReservedGb":38.09476608,"vramReservedPct":44.80815007585195,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":77446.5198586831,"meanTps":77407.6502128878,"stepMs":423.10487365722656,"jitter":0.000200826778098181,"achievedTflops":216.48854442138742,"nominalPeakTflops":989.5,"mfuNominalPct":21.878579527174068,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.878579527174068,"vramAllocatedGb":68.901995008,"vramAllocatedPct":81.04449116082,"vramReservedGb":71.420608512,"vramReservedPct":84.00695617854191,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.736986624,"vramAllocatedPct":98.49383129609701,"vramReservedGb":84.099989504,"vramReservedPct":98.92080563401127,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 262.19 MiB is free. Including non-PyTorch memory, this process has 78.91 GiB memory in use. Of the allocated memory 77.99 GiB is allocated by PyTorch, and 206.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":32241.58585790251,"meanTps":32388.20998645235,"stepMs":63.52044868469238,"jitter":0.01929605404124294,"achievedTflops":90.1258572360748,"nominalPeakTflops":989.5,"mfuNominalPct":9.108222055186943,"configuredPeakTflops":989.5,"mfuConfiguredPct":9.108222055186943,"vramAllocatedGb":8.949571584,"vramAllocatedPct":5.961581613342175,"vramReservedGb":9.032433664,"vramReservedPct":6.0167785630435935,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":57021.518584748854,"meanTps":57008.98630401436,"stepMs":71.8325309753418,"jitter":0.0006329434846198982,"achievedTflops":159.39393508752138,"nominalPeakTflops":989.5,"mfuNominalPct":16.108533106369013,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.108533106369013,"vramAllocatedGb":12.946415104,"vramAllocatedPct":8.624000547767652,"vramReservedGb":13.293846528,"vramReservedPct":8.85543517788097,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":70935.43620488992,"meanTps":70954.82294656071,"stepMs":115.48529815673828,"jitter":0.0007315108210148069,"achievedTflops":198.28791997257255,"nominalPeakTflops":989.5,"mfuNominalPct":20.039203635429264,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.039203635429264,"vramAllocatedGb":20.940069376,"vramAllocatedPct":13.948816588857976,"vramReservedGb":21.53775104,"vramReservedPct":14.346950509045206,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":80034.68884493542,"meanTps":80038.390095824,"stepMs":204.71123504638672,"jitter":0.00034567592297097286,"achievedTflops":223.72332963281772,"nominalPeakTflops":989.5,"mfuNominalPct":22.609735182700124,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.609735182700124,"vramAllocatedGb":36.92737792,"vramAllocatedPct":24.59844867103862,"vramReservedGb":38.09476608,"vramReservedPct":25.376081401831176,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":85664.53600024189,"meanTps":85670.74963071138,"stepMs":382.5153503417969,"jitter":0.0021862390951905186,"achievedTflops":239.46060766921147,"nominalPeakTflops":989.5,"mfuNominalPct":24.200162472886454,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.200162472886454,"vramAllocatedGb":68.901995008,"vramAllocatedPct":45.897712835399915,"vramReservedGb":71.420608512,"vramReservedPct":47.575437832136664,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":88367.40149264068,"meanTps":88343.30733096777,"stepMs":741.6309509277344,"jitter":0.0013436384183320362,"achievedTflops":247.01600741194898,"nominalPeakTflops":989.5,"mfuNominalPct":24.96371979908529,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.96371979908529,"vramAllocatedGb":132.851229184,"vramAllocatedPct":88.4962411641225,"vramReservedGb":137.774497792,"vramReservedPct":91.77578000411235,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.101200896,"vramAllocatedPct":98.65471077453189,"vramReservedGb":148.860043264,"vramReservedPct":99.16019873739502,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.01 GiB is free. Including non-PyTorch memory, this process has 138.79 GiB memory in use. Of the allocated memory 137.93 GiB is allocated by PyTorch, and 143.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":31883.410499445145,"meanTps":32180.996256363316,"stepMs":64.2340316772461,"jitter":0.012695051175973061,"achievedTflops":89.12463907751153,"nominalPeakTflops":989.5,"mfuNominalPct":9.007037804700508,"configuredPeakTflops":989.5,"mfuConfiguredPct":9.007037804700508,"vramAllocatedGb":8.949571584,"vramAllocatedPct":5.961581613342175,"vramReservedGb":9.032433664,"vramReservedPct":6.0167785630435935,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":57442.74066685445,"meanTps":57368.53753263433,"stepMs":71.30578994750977,"jitter":0.0016607918893174946,"achievedTflops":160.57138961484657,"nominalPeakTflops":989.5,"mfuNominalPct":16.227528005542855,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.227528005542855,"vramAllocatedGb":12.946415104,"vramAllocatedPct":8.624000547767652,"vramReservedGb":13.293846528,"vramReservedPct":8.85543517788097,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":71257.10056283418,"meanTps":71257.66803051066,"stepMs":114.96398162841797,"jitter":0.0007836787180512748,"achievedTflops":199.1870778530126,"nominalPeakTflops":989.5,"mfuNominalPct":20.130073557656655,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.130073557656655,"vramAllocatedGb":20.940069376,"vramAllocatedPct":13.948816588857976,"vramReservedGb":21.53775104,"vramReservedPct":14.346950509045206,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":80257.07754155918,"meanTps":80257.47346652775,"stepMs":204.14398956298828,"jitter":0.0003716811408932226,"achievedTflops":224.34497932496265,"nominalPeakTflops":989.5,"mfuNominalPct":22.672559810506584,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.672559810506584,"vramAllocatedGb":36.92737792,"vramAllocatedPct":24.59844867103862,"vramReservedGb":38.09476608,"vramReservedPct":25.376081401831176,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":86120.82802067045,"meanTps":86091.07418585486,"stepMs":380.4886779785156,"jitter":0.0011682688029767433,"achievedTflops":240.7360942309565,"nominalPeakTflops":989.5,"mfuNominalPct":24.32906460141046,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.32906460141046,"vramAllocatedGb":68.901995008,"vramAllocatedPct":45.897712835399915,"vramReservedGb":71.420608512,"vramReservedPct":47.575437832136664,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":88356.23960057656,"meanTps":88363.18291126395,"stepMs":741.7246398925781,"jitter":0.001315164601213265,"achievedTflops":246.9848062453845,"nominalPeakTflops":989.5,"mfuNominalPct":24.96056657356084,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.96056657356084,"vramAllocatedGb":132.851229184,"vramAllocatedPct":88.4962411641225,"vramReservedGb":137.774497792,"vramReservedPct":91.77578000411235,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.101200896,"vramAllocatedPct":98.65471077453189,"vramReservedGb":148.860043264,"vramReservedPct":99.16019873739502,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.01 GiB is free. Including non-PyTorch memory, this process has 138.79 GiB memory in use. Of the allocated memory 137.93 GiB is allocated by PyTorch, and 143.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":21391.675912058356,"meanTps":21331.14510837349,"stepMs":95.73817443847656,"jitter":0.010601600513969932,"achievedTflops":59.79678350151646,"nominalPeakTflops":364.2,"mfuNominalPct":16.418666529795843,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":13.24597734529423,"vramAllocatedGb":8.89950208,"vramAllocatedPct":17.492245455798454,"vramReservedGb":8.969519104,"vramReservedPct":17.62986607309624,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":27434.20120860106,"meanTps":27496.95984382806,"stepMs":149.3026885986328,"jitter":0.004483013318444784,"achievedTflops":76.68763293496946,"nominalPeakTflops":364.2,"mfuNominalPct":21.056461541726925,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.987580084388387,"vramAllocatedGb":12.8963456,"vramAllocatedPct":25.34816450292985,"vramReservedGb":13.230931968,"vramReservedPct":26.005804314978764,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":27458.650958928603,"meanTps":27473.293797723712,"stepMs":298.33949279785156,"jitter":0.0025661173087509256,"achievedTflops":76.75597804420786,"nominalPeakTflops":364.2,"mfuNominalPct":21.07522735974955,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.00271965738249,"vramAllocatedGb":20.889999872,"vramAllocatedPct":41.05993819067935,"vramReservedGb":21.495808,"vramReservedPct":42.25067272603144,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25862.266551374476,"meanTps":25880.34302755979,"stepMs":633.5098266601562,"jitter":0.003233118376017939,"achievedTflops":72.29355755896194,"nominalPeakTflops":364.2,"mfuNominalPct":19.849960889336064,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.01421965468175,"vramAllocatedGb":36.877308416,"vramAllocatedPct":72.48348556617833,"vramReservedGb":38.05282304,"vramReservedPct":74.79399576720395,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":20811.275616573745,"meanTps":20875.714333986052,"stepMs":98.4081916809082,"jitter":0.01899233193383586,"achievedTflops":58.174373412845206,"nominalPeakTflops":364.2,"mfuNominalPct":15.973194237464362,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":12.886586655345646,"vramAllocatedGb":8.89950208,"vramAllocatedPct":17.492245455798454,"vramReservedGb":8.969519104,"vramReservedPct":17.62986607309624,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":27313.72126650957,"meanTps":27286.34818549066,"stepMs":149.9612579345703,"jitter":0.0050239851581132455,"achievedTflops":76.35085179434552,"nominalPeakTflops":364.2,"mfuNominalPct":20.963990058853796,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.912977487094608,"vramAllocatedGb":12.8963456,"vramAllocatedPct":25.34816450292985,"vramReservedGb":13.230931968,"vramReservedPct":26.005804314978764,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":27376.34540305146,"meanTps":27370.787157051178,"stepMs":299.23643493652344,"jitter":0.0020595321782293074,"achievedTflops":76.52590689288758,"nominalPeakTflops":364.2,"mfuNominalPct":21.012055709194833,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.951755089060565,"vramAllocatedGb":20.889999872,"vramAllocatedPct":41.05993819067935,"vramReservedGb":21.495808,"vramReservedPct":42.25067272603144,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25878.553739952476,"meanTps":25867.993962760538,"stepMs":633.1111145019531,"jitter":0.003854542674572208,"achievedTflops":72.33908561825243,"nominalPeakTflops":364.2,"mfuNominalPct":19.862461729338943,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.024304873428036,"vramAllocatedGb":36.877308416,"vramAllocatedPct":72.48348556617833,"vramReservedGb":38.05282304,"vramReservedPct":74.79399576720395,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":12754.892190256947,"meanTps":12742.477020442486,"stepMs":160.5658416748047,"jitter":0.0031972253601046784,"achievedTflops":35.65412686792098,"nominalPeakTflops":154.8,"mfuNominalPct":23.032381697623368,"configuredPeakTflops":180.6,"mfuConfiguredPct":19.742041455105745,"vramAllocatedGb":8.89950208,"vramAllocatedPct":17.438401805022625,"vramReservedGb":8.969519104,"vramReservedPct":17.575598806240016,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":15365.070911826846,"meanTps":15366.299301102043,"stepMs":266.5786590576172,"jitter":0.0011986907185650506,"achievedTflops":42.950436542564034,"nominalPeakTflops":154.8,"mfuNominalPct":27.745760040416044,"configuredPeakTflops":180.6,"mfuConfiguredPct":23.782080034642327,"vramAllocatedGb":12.8963456,"vramAllocatedPct":25.27013920190416,"vramReservedGb":13.230931968,"vramReservedPct":25.925754703897184,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":17307.477074006834,"meanTps":17314.241493980913,"stepMs":473.3214416503906,"jitter":0.0009234587209619829,"achievedTflops":48.38010185861416,"nominalPeakTflops":154.8,"mfuNominalPct":31.25329577429855,"configuredPeakTflops":180.6,"mfuConfiguredPct":26.788539235113046,"vramAllocatedGb":20.889999872,"vramAllocatedPct":40.93354978740645,"vramReservedGb":21.495808,"vramReservedPct":42.12061907036712,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":18601.168711450897,"meanTps":18604.505985190088,"stepMs":880.8048706054688,"jitter":0.0009624859756081081,"achievedTflops":51.99639630323775,"nominalPeakTflops":154.8,"mfuNominalPct":33.58940329666521,"configuredPeakTflops":180.6,"mfuConfiguredPct":28.790917111427326,"vramAllocatedGb":36.877308416,"vramAllocatedPct":72.26037095841103,"vramReservedGb":38.05282304,"vramReservedPct":74.56376907627428,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.538287104,"vramAllocatedPct":99.02879387350458,"vramReservedGb":50.671386624,"vramReservedPct":99.28959980275222,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 68.25 MiB is free. Process 1348507 has 47.46 GiB memory in use. Of the allocated memory 47.00 GiB is allocated by PyTorch, and 130.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":12697.039738471743,"meanTps":12702.765410070086,"stepMs":161.2974395751953,"jitter":0.0026186904875312523,"achievedTflops":35.49241020071578,"nominalPeakTflops":154.8,"mfuNominalPct":22.927913566353862,"configuredPeakTflops":180.6,"mfuConfiguredPct":19.652497342589026,"vramAllocatedGb":8.89950208,"vramAllocatedPct":17.438401805022625,"vramReservedGb":8.969519104,"vramReservedPct":17.575598806240016,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":15281.704466051828,"meanTps":15277.951733137497,"stepMs":268.0329284667969,"jitter":0.0012623255431603879,"achievedTflops":42.71739985437778,"nominalPeakTflops":154.8,"mfuNominalPct":27.59521954417169,"configuredPeakTflops":180.6,"mfuConfiguredPct":23.65304532357574,"vramAllocatedGb":12.8963456,"vramAllocatedPct":25.27013920190416,"vramReservedGb":13.230931968,"vramReservedPct":25.925754703897184,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":17213.277263337557,"meanTps":17212.869035540767,"stepMs":475.9116973876953,"jitter":0.0012979081544792744,"achievedTflops":48.11678234556473,"nominalPeakTflops":154.8,"mfuNominalPct":31.08319272969298,"configuredPeakTflops":180.6,"mfuConfiguredPct":26.64273662545113,"vramAllocatedGb":20.889999872,"vramAllocatedPct":40.93354978740645,"vramReservedGb":21.495808,"vramReservedPct":42.12061907036712,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":18577.316777298638,"meanTps":18574.509127644695,"stepMs":881.9357604980469,"jitter":0.0005187178804614137,"achievedTflops":51.92972228721122,"nominalPeakTflops":154.8,"mfuNominalPct":33.54633222688063,"configuredPeakTflops":180.6,"mfuConfiguredPct":28.75399905161197,"vramAllocatedGb":36.877308416,"vramAllocatedPct":72.26037095841103,"vramReservedGb":38.05282304,"vramReservedPct":74.56376907627428,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.538287104,"vramAllocatedPct":99.02879387350458,"vramReservedGb":50.671386624,"vramReservedPct":99.28959980275222,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 68.25 MiB is free. Process 1380474 has 47.46 GiB memory in use. Of the allocated memory 47.00 GiB is allocated by PyTorch, and 130.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":34283.32987650436,"meanTps":34287.58729977585,"stepMs":59.73748779296875,"jitter":0.0006027982306825415,"achievedTflops":95.83320459622387,"nominalPeakTflops":468.0,"mfuNominalPct":20.4771804692786,"configuredPeakTflops":468.0,"mfuConfiguredPct":20.4771804692786,"vramAllocatedGb":8.89950208,"vramAllocatedPct":8.727230865829606,"vramReservedGb":8.969519104,"vramReservedPct":8.795892542347392,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":45561.28209893673,"meanTps":45504.722154738636,"stepMs":89.90089416503906,"jitter":0.0005599623965395651,"achievedTflops":127.35879754918592,"nominalPeakTflops":468.0,"mfuNominalPct":27.213418279740583,"configuredPeakTflops":468.0,"mfuConfiguredPct":27.213418279740583,"vramAllocatedGb":12.8963456,"vramAllocatedPct":12.646705890395816,"vramReservedGb":13.230931968,"vramReservedPct":12.974815536513841,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":53494.79080076893,"meanTps":53492.22997211872,"stepMs":153.13640594482422,"jitter":0.00027928158253119595,"achievedTflops":149.5355687475305,"nominalPeakTflops":468.0,"mfuNominalPct":31.952044604173185,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.952044604173185,"vramAllocatedGb":20.889999872,"vramAllocatedPct":20.485623805831494,"vramReservedGb":21.495808,"vramReservedPct":21.079705064077803,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":54213.0773187621,"meanTps":54209.1275375978,"stepMs":302.21490478515625,"jitter":0.00014462231462091503,"achievedTflops":151.54341626658737,"nominalPeakTflops":468.0,"mfuNominalPct":32.38107185183491,"configuredPeakTflops":468.0,"mfuConfiguredPct":32.38107185183491,"vramAllocatedGb":36.877308416,"vramAllocatedPct":36.163459636702854,"vramReservedGb":38.05282304,"vramReservedPct":37.31621935489676,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":49363.36628770178,"meanTps":49353.89996981563,"stepMs":663.8121032714844,"jitter":0.00019480097159797614,"achievedTflops":137.98687578039946,"nominalPeakTflops":468.0,"mfuNominalPct":29.48437516675202,"configuredPeakTflops":468.0,"mfuConfiguredPct":29.48437516675202,"vramAllocatedGb":68.851925504,"vramAllocatedPct":67.51913129844556,"vramReservedGb":71.420608512,"vramReservedPct":70.03809128412037,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":101.0605312,"vramAllocatedPct":99.10426215730216,"vramReservedGb":101.233721344,"vramReservedPct":99.27409979055257,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 64.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 35.75 MiB is free. Including non-PyTorch memory, this process has 94.93 GiB memory in use. Of the allocated memory 94.12 GiB is allocated by PyTorch, and 165.17 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":34195.248825792296,"meanTps":34184.382875899035,"stepMs":59.891361236572266,"jitter":0.0010159483141606,"achievedTflops":95.58698903360651,"nominalPeakTflops":468.0,"mfuNominalPct":20.424570306326178,"configuredPeakTflops":468.0,"mfuConfiguredPct":20.424570306326178,"vramAllocatedGb":8.89950208,"vramAllocatedPct":8.727230865829606,"vramReservedGb":8.969519104,"vramReservedPct":8.795892542347392,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":45490.39909078669,"meanTps":45489.43779532667,"stepMs":90.04097747802734,"jitter":0.0005448328571509673,"achievedTflops":127.16065618290358,"nominalPeakTflops":468.0,"mfuNominalPct":27.171080380962305,"configuredPeakTflops":468.0,"mfuConfiguredPct":27.171080380962305,"vramAllocatedGb":12.8963456,"vramAllocatedPct":12.646705890395816,"vramReservedGb":13.230931968,"vramReservedPct":12.974815536513841,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":53472.45810043068,"meanTps":53467.41497881803,"stepMs":153.2003631591797,"jitter":0.00023228062878412337,"achievedTflops":149.47314149065633,"nominalPeakTflops":468.0,"mfuNominalPct":31.938705446721436,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.938705446721436,"vramAllocatedGb":20.889999872,"vramAllocatedPct":20.485623805831494,"vramReservedGb":21.495808,"vramReservedPct":21.079705064077803,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":54189.12636117309,"meanTps":54190.99374204957,"stepMs":302.3484802246094,"jitter":0.00032809972994810423,"achievedTflops":151.47646544742335,"nominalPeakTflops":468.0,"mfuNominalPct":32.366766121244304,"configuredPeakTflops":468.0,"mfuConfiguredPct":32.366766121244304,"vramAllocatedGb":36.877308416,"vramAllocatedPct":36.163459636702854,"vramReservedGb":38.05282304,"vramReservedPct":37.31621935489676,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":49341.65321507606,"meanTps":49342.16587881199,"stepMs":664.1042175292969,"jitter":0.00021222467320182597,"achievedTflops":137.92618058717147,"nominalPeakTflops":468.0,"mfuNominalPct":29.471406108369973,"configuredPeakTflops":468.0,"mfuConfiguredPct":29.471406108369973,"vramAllocatedGb":68.851925504,"vramAllocatedPct":67.51913129844556,"vramReservedGb":71.420608512,"vramReservedPct":70.03809128412037,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":101.0605312,"vramAllocatedPct":99.10426215730216,"vramReservedGb":101.233721344,"vramReservedPct":99.27409979055257,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 64.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 35.75 MiB is free. Including non-PyTorch memory, this process has 94.93 GiB memory in use. Of the allocated memory 94.12 GiB is allocated by PyTorch, and 165.17 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":31581.23821043915,"meanTps":31576.89166210201,"stepMs":518.7890319824219,"jitter":0.0008503307621279201,"achievedTflops":110.53346866325997,"nominalPeakTflops":312.0,"mfuNominalPct":35.42739380232691,"configuredPeakTflops":312.0,"mfuConfiguredPct":35.42739380232691,"vramAllocatedGb":36.885928448,"vramAllocatedPct":43.40836535769595,"vramReservedGb":38.038142976,"vramReservedPct":44.76432280017646,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":32951.017521359405,"meanTps":32949.47540982409,"stepMs":994.4457702636719,"jitter":0.0014234525570322395,"achievedTflops":115.32765873048596,"nominalPeakTflops":312.0,"mfuNominalPct":36.963993182848064,"configuredPeakTflops":312.0,"mfuConfiguredPct":36.963993182848064,"vramAllocatedGb":68.861825536,"vramAllocatedPct":81.03847206336707,"vramReservedGb":71.399636992,"vramReservedPct":84.02503771390494,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.694257152,"vramAllocatedPct":98.49368161944588,"vramReservedGb":84.13773824,"vramReservedPct":99.0155822440776,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 512.75 MiB is free. Process 764857 has 78.63 GiB memory in use. Of the allocated memory 77.95 GiB is allocated by PyTorch, and 182.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":31511.16717312847,"meanTps":31511.513592091218,"stepMs":519.9426574707031,"jitter":0.0015392081024184266,"achievedTflops":110.28822195205844,"nominalPeakTflops":312.0,"mfuNominalPct":35.34878908719821,"configuredPeakTflops":312.0,"mfuConfiguredPct":35.34878908719821,"vramAllocatedGb":36.885928448,"vramAllocatedPct":43.40836535769595,"vramReservedGb":38.038142976,"vramReservedPct":44.76432280017646,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":32907.47111782359,"meanTps":32917.080985636545,"stepMs":995.76171875,"jitter":0.001273580022997862,"achievedTflops":115.1752475109337,"nominalPeakTflops":312.0,"mfuNominalPct":36.91514343299157,"configuredPeakTflops":312.0,"mfuConfiguredPct":36.91514343299157,"vramAllocatedGb":68.861825536,"vramAllocatedPct":81.03847206336707,"vramReservedGb":71.399636992,"vramReservedPct":84.02503771390494,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.694257152,"vramAllocatedPct":98.49368161944588,"vramReservedGb":84.13773824,"vramReservedPct":99.0155822440776,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 512.75 MiB is free. Process 852916 has 78.63 GiB memory in use. Of the allocated memory 77.95 GiB is allocated by PyTorch, and 182.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":94373.74175992305,"meanTps":94355.7945391345,"stepMs":347.21522521972656,"jitter":0.0006796279796126556,"achievedTflops":330.30551107419643,"nominalPeakTflops":2250.0,"mfuNominalPct":14.68024493663095,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.68024493663095,"vramAllocatedGb":68.861825536,"vramAllocatedPct":35.95861305116108,"vramReservedGb":71.290585088,"vramReservedPct":37.226874881934386,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":100528.81828562677,"meanTps":100508.20837950078,"stepMs":651.9125671386719,"jitter":0.0001724690220252711,"achievedTflops":351.8481103143032,"nominalPeakTflops":2250.0,"mfuNominalPct":15.637693791746809,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.637693791746809,"vramAllocatedGb":132.813619712,"vramAllocatedPct":69.35328132785486,"vramReservedGb":137.690611712,"vramReservedPct":71.89997344379256,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.422552576,"vramAllocatedPct":99.4358025073714,"vramReservedGb":190.551425024,"vramReservedPct":99.5030977679097,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 155.81 MiB is free. Including non-PyTorch memory, this process has 178.19 GiB memory in use. Of the allocated memory 177.24 GiB is allocated by PyTorch, and 134.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":94104.46477048195,"meanTps":93934.88751446678,"stepMs":348.2087707519531,"jitter":0.002368629859234424,"achievedTflops":329.3630489871879,"nominalPeakTflops":2250.0,"mfuNominalPct":14.638357732763906,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.638357732763906,"vramAllocatedGb":68.861825536,"vramAllocatedPct":35.95861305116108,"vramReservedGb":71.290585088,"vramReservedPct":37.226874881934386,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":100459.9650307777,"meanTps":100389.15799309425,"stepMs":652.359375,"jitter":0.0007718684601117339,"achievedTflops":351.60712580835985,"nominalPeakTflops":2250.0,"mfuNominalPct":15.626983369260438,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.626983369260438,"vramAllocatedGb":132.813619712,"vramAllocatedPct":69.35328132785486,"vramReservedGb":137.690611712,"vramReservedPct":71.89997344379256,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.422552576,"vramAllocatedPct":99.4358025073714,"vramReservedGb":190.551425024,"vramReservedPct":99.5030977679097,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 155.81 MiB is free. Including non-PyTorch memory, this process has 178.19 GiB memory in use. Of the allocated memory 177.24 GiB is allocated by PyTorch, and 134.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":100884.41888300108,"meanTps":100857.64632060872,"stepMs":649.6146850585938,"jitter":0.0003024182848775947,"achievedTflops":353.09270266450176,"nominalPeakTflops":2250.0,"mfuNominalPct":15.69300900731119,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.69300900731119,"vramAllocatedGb":132.813619712,"vramAllocatedPct":46.20751062146289,"vramReservedGb":137.652862976,"vramReservedPct":47.89114355764826,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":32,"tokensPerStep":131072,"status":"complete","stable":true,"tps":105128.62961499229,"meanTps":105123.08866840988,"stepMs":1246.7774047851562,"jitter":0.0001864572341555752,"achievedTflops":367.94732396905056,"nominalPeakTflops":2250.0,"mfuNominalPct":16.35321439862447,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.35321439862447,"vramAllocatedGb":260.717208064,"vramAllocatedPct":90.70676024747293,"vramReservedGb":270.28094976,"vramReservedPct":94.03410496525957,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.99077888,"vramAllocatedPct":99.49974996083964,"vramReservedGb":286.517100544,"vramReservedPct":99.68286381567113,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 435.62 MiB is free. Including non-PyTorch memory, this process has 267.25 GiB memory in use. Of the allocated memory 266.35 GiB is allocated by PyTorch, and 81.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":101011.13732115953,"meanTps":100991.41442287547,"stepMs":648.7997436523438,"jitter":0.0004609617086087544,"achievedTflops":353.5362137269849,"nominalPeakTflops":2250.0,"mfuNominalPct":15.712720610088216,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.712720610088216,"vramAllocatedGb":132.813619712,"vramAllocatedPct":46.20751062146289,"vramReservedGb":137.652862976,"vramReservedPct":47.89114355764826,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":32,"tokensPerStep":131072,"status":"complete","stable":true,"tps":105181.42319802157,"meanTps":105183.22889012237,"stepMs":1246.151611328125,"jitter":8.38716442829751e-05,"achievedTflops":368.13210006353125,"nominalPeakTflops":2250.0,"mfuNominalPct":16.361426669490278,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.361426669490278,"vramAllocatedGb":260.717208064,"vramAllocatedPct":90.70676024747293,"vramReservedGb":270.28094976,"vramReservedPct":94.03410496525957,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.99077888,"vramAllocatedPct":99.49974996083964,"vramReservedGb":286.517100544,"vramReservedPct":99.68286381567113,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 435.62 MiB is free. Including non-PyTorch memory, this process has 267.25 GiB memory in use. Of the allocated memory 266.35 GiB is allocated by PyTorch, and 81.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":26282.714524392697,"meanTps":26251.992905113544,"stepMs":155.8438720703125,"jitter":0.0014591731007078555,"achievedTflops":91.98878089925819,"nominalPeakTflops":165.2,"mfuNominalPct":55.68328141601585,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":44.75976687985963,"vramAllocatedGb":12.903972864,"vramAllocatedPct":51.08249254772956,"vramReservedGb":12.952010752,"vramReservedPct":51.272658404526624,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":26844.68515688491,"meanTps":26846.626787356385,"stepMs":305.16282653808594,"jitter":0.00642274758563501,"achievedTflops":93.95566271948135,"nominalPeakTflops":165.2,"mfuNominalPct":56.87388784472237,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":45.7168093680065,"vramAllocatedGb":20.897979904,"vramAllocatedPct":82.72808025556942,"vramReservedGb":21.50629376,"vramReservedPct":85.136190404537,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.63570432,"vramAllocatedPct":97.52447526027817,"vramReservedGb":24.708644864,"vramReservedPct":97.81322236433495,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 88.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 47.69 MiB is free. Including non-PyTorch memory, this process has 23.47 GiB memory in use. Of the allocated memory 22.94 GiB is allocated by PyTorch, and 69.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":26246.88564733343,"meanTps":26246.98855832318,"stepMs":156.05661010742188,"jitter":0.0008802974012953141,"achievedTflops":91.86338081097536,"nominalPeakTflops":165.2,"mfuNominalPct":55.607373372261115,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":44.698749887750566,"vramAllocatedGb":12.903972864,"vramAllocatedPct":51.08249254772956,"vramReservedGb":12.952010752,"vramReservedPct":51.272658404526624,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":26975.565365677896,"meanTps":26972.18498587963,"stepMs":303.68223571777344,"jitter":0.0007660277681989541,"achievedTflops":94.41373986518614,"nominalPeakTflops":165.2,"mfuNominalPct":57.15117425253399,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":45.93969987763498,"vramAllocatedGb":20.897979904,"vramAllocatedPct":82.72808025556942,"vramReservedGb":21.50629376,"vramReservedPct":85.136190404537,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.63570432,"vramAllocatedPct":97.52447526027817,"vramReservedGb":24.708644864,"vramReservedPct":97.81322236433495,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 88.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 47.69 MiB is free. Including non-PyTorch memory, this process has 23.47 GiB memory in use. Of the allocated memory 22.94 GiB is allocated by PyTorch, and 69.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":34369.534510204234,"meanTps":34373.06708848186,"stepMs":119.17531204223633,"jitter":0.0005248713929439025,"achievedTflops":120.29242933542551,"nominalPeakTflops":209.5,"mfuNominalPct":57.41882068516731,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":44.72721727805751,"vramAllocatedGb":12.903972864,"vramAllocatedPct":38.32613781907791,"vramReservedGb":12.952010752,"vramReservedPct":38.4688153289758,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":37852.814234952224,"meanTps":37846.251568379994,"stepMs":216.41719818115234,"jitter":0.00040741631732233764,"achievedTflops":132.48381295804526,"nominalPeakTflops":209.5,"mfuNominalPct":63.23809687734857,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":49.26022627306732,"vramAllocatedGb":20.897979904,"vramAllocatedPct":62.06916787429975,"vramReservedGb":21.48532224,"vramReservedPct":63.81363553195548,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.692646912,"vramAllocatedPct":97.10055224274252,"vramReservedGb":32.8204288,"vramReservedPct":97.480077703768,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 239.88 MiB is free. Including non-PyTorch memory, this process has 31.11 GiB memory in use. Of the allocated memory 30.45 GiB is allocated by PyTorch, and 81.86 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":34376.11129729757,"meanTps":34376.19659088943,"stepMs":119.15251159667969,"jitter":0.00045491229806862243,"achievedTflops":120.31544791010083,"nominalPeakTflops":209.5,"mfuNominalPct":57.42980807164717,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":44.73577606098859,"vramAllocatedGb":12.903972864,"vramAllocatedPct":38.32613781907791,"vramReservedGb":12.952010752,"vramReservedPct":38.4688153289758,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":37837.63848769234,"meanTps":37833.74320406103,"stepMs":216.50399780273438,"jitter":0.0002765349600241204,"achievedTflops":132.43069825832973,"nominalPeakTflops":209.5,"mfuNominalPct":63.212743798725405,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":49.24047712735665,"vramAllocatedGb":20.897979904,"vramAllocatedPct":62.06916787429975,"vramReservedGb":21.48532224,"vramReservedPct":63.81363553195548,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.692646912,"vramAllocatedPct":97.10055224274252,"vramReservedGb":32.8204288,"vramReservedPct":97.480077703768,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 239.88 MiB is free. Including non-PyTorch memory, this process has 31.11 GiB memory in use. Of the allocated memory 30.45 GiB is allocated by PyTorch, and 81.86 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":65340.83493413258,"meanTps":65310.71331736398,"stepMs":250.74671936035156,"jitter":0.0005289671468952871,"achievedTflops":228.69113245331354,"nominalPeakTflops":989.5,"mfuNominalPct":23.111787008925067,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.111787008925067,"vramAllocatedGb":36.935997952,"vramAllocatedPct":43.445173963241096,"vramReservedGb":38.101057536,"vramReservedPct":44.81555026579016,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":69791.71193141869,"meanTps":69784.54544010838,"stepMs":469.51133728027344,"jitter":0.0004414086755902157,"achievedTflops":244.26908002539216,"nominalPeakTflops":989.5,"mfuNominalPct":24.686112180433767,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.686112180433767,"vramAllocatedGb":68.91189504,"vramAllocatedPct":81.0561358607423,"vramReservedGb":71.357693952,"vramReservedPct":83.93295427915983,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.744326656,"vramAllocatedPct":98.50246485102492,"vramReservedGb":84.17968128,"vramReservedPct":99.01454137322858,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 308.19 MiB is free. Including non-PyTorch memory, this process has 78.87 GiB memory in use. Of the allocated memory 77.99 GiB is allocated by PyTorch, and 155.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":65359.42702400817,"meanTps":65343.630330483364,"stepMs":250.6753921508789,"jitter":0.0007871335876207238,"achievedTflops":228.75620425860353,"nominalPeakTflops":989.5,"mfuNominalPct":23.118363239879084,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.118363239879084,"vramAllocatedGb":36.935997952,"vramAllocatedPct":43.445173963241096,"vramReservedGb":38.101057536,"vramReservedPct":44.81555026579016,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":69797.3896520056,"meanTps":69794.21405232359,"stepMs":469.47314453125,"jitter":0.00040202730865196505,"achievedTflops":244.2889518919223,"nominalPeakTflops":989.5,"mfuNominalPct":24.688120453958796,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.688120453958796,"vramAllocatedGb":68.91189504,"vramAllocatedPct":81.0561358607423,"vramReservedGb":71.357693952,"vramReservedPct":83.93295427915983,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.744326656,"vramAllocatedPct":98.50246485102492,"vramReservedGb":84.17968128,"vramReservedPct":99.01454137322858,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 308.19 MiB is free. Including non-PyTorch memory, this process has 78.87 GiB memory in use. Of the allocated memory 77.99 GiB is allocated by PyTorch, and 155.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":76367.57581935565,"meanTps":76355.38328022258,"stepMs":429.08262634277344,"jitter":0.0011587541538854464,"achievedTflops":267.28442350710793,"nominalPeakTflops":989.5,"mfuNominalPct":27.012069076008885,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.012069076008885,"vramAllocatedGb":68.91189504,"vramAllocatedPct":45.90430754758125,"vramReservedGb":71.374471168,"vramReservedPct":47.544704345165,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":78433.63990558672,"meanTps":78390.78120605713,"stepMs":835.5598449707031,"jitter":0.0022896682624396793,"achievedTflops":274.51559121528925,"nominalPeakTflops":989.5,"mfuNominalPct":27.74285914252544,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.74285914252544,"vramAllocatedGb":132.863689216,"vramAllocatedPct":88.50454117010331,"vramReservedGb":137.774497792,"vramReservedPct":91.77578000411235,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.108540928,"vramAllocatedPct":98.65960019291374,"vramReservedGb":149.036204032,"vramReservedPct":99.27754477855954,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.04 GiB is free. Including non-PyTorch memory, this process has 138.76 GiB memory in use. Of the allocated memory 137.94 GiB is allocated by PyTorch, and 104.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":76338.00472158153,"meanTps":76360.0873583119,"stepMs":429.24884033203125,"jitter":0.0012441585678447586,"achievedTflops":267.18092547491005,"nominalPeakTflops":989.5,"mfuNominalPct":27.001609446681154,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.001609446681154,"vramAllocatedGb":68.91189504,"vramAllocatedPct":45.90430754758125,"vramReservedGb":71.374471168,"vramReservedPct":47.544704345165,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":78476.93179062041,"meanTps":78427.41661343665,"stepMs":835.0989074707031,"jitter":0.001959460996960261,"achievedTflops":274.6671116270558,"nominalPeakTflops":989.5,"mfuNominalPct":27.758171968373503,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.758171968373503,"vramAllocatedGb":132.863689216,"vramAllocatedPct":88.50454117010331,"vramReservedGb":137.774497792,"vramReservedPct":91.77578000411235,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.108540928,"vramAllocatedPct":98.65960019291374,"vramReservedGb":149.036204032,"vramReservedPct":99.27754477855954,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.04 GiB is free. Including non-PyTorch memory, this process has 138.76 GiB memory in use. Of the allocated memory 137.94 GiB is allocated by PyTorch, and 104.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":25643.272484299985,"meanTps":25626.094564763116,"stepMs":159.7300033569336,"jitter":0.003848320007168468,"achievedTflops":89.75075127453006,"nominalPeakTflops":364.2,"mfuNominalPct":24.643259548196067,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":19.88127702680552,"vramAllocatedGb":12.903972864,"vramAllocatedPct":25.363156125252633,"vramReservedGb":12.952010752,"vramReservedPct":25.457576073753188,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25269.43658670442,"meanTps":25283.930969951256,"stepMs":324.18609619140625,"jitter":0.004374347802348316,"achievedTflops":88.44233587305848,"nominalPeakTflops":364.2,"mfuNominalPct":24.284002161740382,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":19.591441357539342,"vramAllocatedGb":20.897979904,"vramAllocatedPct":41.075623189371896,"vramReservedGb":21.50629376,"vramReservedPct":42.27128281028804,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":24065.31453420694,"meanTps":24056.95263592564,"stepMs":680.8138732910156,"jitter":0.0033250808365393393,"achievedTflops":84.22794167262856,"nominalPeakTflops":364.2,"mfuNominalPct":23.12683736206166,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":18.65788328243622,"vramAllocatedGb":36.885928448,"vramAllocatedPct":72.50042850458381,"vramReservedGb":38.0633088,"vramReservedPct":74.81460585146054,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":16057.600648536098,"meanTps":16050.843455935563,"stepMs":510.16339111328125,"jitter":0.0012359113644202673,"achievedTflops":56.201162420079385,"nominalPeakTflops":154.8,"mfuNominalPct":36.305660478087454,"configuredPeakTflops":180.6,"mfuConfiguredPct":31.119137552646393,"vramAllocatedGb":20.897979904,"vramAllocatedPct":40.94918650541404,"vramReservedGb":21.500002304,"vramReservedPct":42.1288377277467,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":17278.954727844455,"meanTps":17273.580541819952,"stepMs":948.2055053710938,"jitter":0.0003994375876667216,"achievedTflops":60.475868242327685,"nominalPeakTflops":154.8,"mfuNominalPct":39.067098347756904,"configuredPeakTflops":180.6,"mfuConfiguredPct":33.486084298077344,"vramAllocatedGb":36.885928448,"vramAllocatedPct":72.27726174401194,"vramReservedGb":38.0633088,"vramReservedPct":74.58431571972324,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.545627136,"vramAllocatedPct":99.04317652391886,"vramReservedGb":50.671386624,"vramReservedPct":99.28959980275222,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 68.25 MiB is free. Process 1415479 has 47.46 GiB memory in use. Of the allocated memory 47.01 GiB is allocated by PyTorch, and 123.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":16129.411455789348,"meanTps":16130.32349044861,"stepMs":507.8920593261719,"jitter":0.0010846940987070113,"achievedTflops":56.45249827842412,"nominalPeakTflops":154.8,"mfuNominalPct":36.468022143684834,"configuredPeakTflops":180.6,"mfuConfiguredPct":31.258304694587004,"vramAllocatedGb":20.897979904,"vramAllocatedPct":40.94918650541404,"vramReservedGb":21.500002304,"vramReservedPct":42.1288377277467,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":17299.73001743117,"meanTps":17299.59466873973,"stepMs":947.0668029785156,"jitter":0.00046646731425748666,"achievedTflops":60.54858118680446,"nominalPeakTflops":154.8,"mfuNominalPct":39.114070534111406,"configuredPeakTflops":180.6,"mfuConfiguredPct":33.526346172095494,"vramAllocatedGb":36.885928448,"vramAllocatedPct":72.27726174401194,"vramReservedGb":38.0633088,"vramReservedPct":74.58431571972324,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.545627136,"vramAllocatedPct":99.04317652391886,"vramReservedGb":50.671386624,"vramReservedPct":99.28959980275222,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 68.25 MiB is free. Process 1449495 has 47.46 GiB memory in use. Of the allocated memory 47.01 GiB is allocated by PyTorch, and 123.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":49771.69619802068,"meanTps":49770.07090440822,"stepMs":164.59153747558594,"jitter":0.0003575694391717333,"achievedTflops":174.19957334677017,"nominalPeakTflops":468.0,"mfuNominalPct":37.22213105700217,"configuredPeakTflops":468.0,"mfuConfiguredPct":37.22213105700217,"vramAllocatedGb":20.897979904,"vramAllocatedPct":20.493449365166686,"vramReservedGb":21.500002304,"vramReservedPct":21.08381817726104,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":50740.43914075687,"meanTps":50735.7257911761,"stepMs":322.8982696533203,"jitter":0.00034610294629468165,"achievedTflops":177.5901471105401,"nominalPeakTflops":468.0,"mfuNominalPct":37.94661263045729,"configuredPeakTflops":468.0,"mfuConfiguredPct":37.94661263045729,"vramAllocatedGb":36.885928448,"vramAllocatedPct":36.171912807302576,"vramReservedGb":38.0633088,"vramReservedPct":37.326502137854845,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":46591.90573432158,"meanTps":46584.19475414198,"stepMs":703.2981262207031,"jitter":0.00040040932629739216,"achievedTflops":163.07039382464365,"nominalPeakTflops":468.0,"mfuNominalPct":34.84410124458198,"configuredPeakTflops":468.0,"mfuConfiguredPct":34.84410124458198,"vramAllocatedGb":68.861825536,"vramAllocatedPct":67.52883969157435,"vramReservedGb":71.4080256,"vramReservedPct":70.02575194457066,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":49531.659840125576,"meanTps":49520.0176193449,"stepMs":165.38916778564453,"jitter":0.00024655412711080865,"achievedTflops":173.3594526692132,"nominalPeakTflops":468.0,"mfuNominalPct":37.04261809171222,"configuredPeakTflops":468.0,"mfuConfiguredPct":37.04261809171222,"vramAllocatedGb":21.018809856,"vramAllocatedPct":20.61194036355422,"vramReservedGb":22.735224832,"vramReservedPct":22.295130009723657,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":50782.41813457152,"meanTps":50739.52056947626,"stepMs":322.63134765625,"jitter":0.0008095973092656968,"achievedTflops":177.7370724390028,"nominalPeakTflops":468.0,"mfuNominalPct":37.97800693141085,"configuredPeakTflops":468.0,"mfuConfiguredPct":37.97800693141085,"vramAllocatedGb":36.990493184,"vramAllocatedPct":36.274453441968795,"vramReservedGb":38.887489536,"vramReservedPct":38.13472887836046,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":46569.90161846332,"meanTps":46564.58991900241,"stepMs":703.6304321289062,"jitter":0.0003282011602597126,"achievedTflops":162.9933800218765,"nominalPeakTflops":468.0,"mfuNominalPct":34.827645303819764,"configuredPeakTflops":468.0,"mfuConfiguredPct":34.827645303819764,"vramAllocatedGb":68.995981824,"vramAllocatedPct":67.66039906275647,"vramReservedGb":71.791804416,"vramReservedPct":70.40210180083663,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":26808.35025736462,"meanTps":26799.204038985154,"stepMs":611.1528625488281,"jitter":0.0016226510425787503,"achievedTflops":131.60912812764872,"nominalPeakTflops":312.0,"mfuNominalPct":42.18241286142587,"configuredPeakTflops":312.0,"mfuConfiguredPct":42.18241286142587,"vramAllocatedGb":36.901248512,"vramAllocatedPct":43.42639442632446,"vramReservedGb":38.05282304,"vramReservedPct":44.78159869937159,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":27895.582875205786,"meanTps":27893.11875066639,"stepMs":1174.666259765625,"jitter":0.0013826288758463435,"achievedTflops":136.94663437224534,"nominalPeakTflops":312.0,"mfuNominalPct":43.893152042386326,"configuredPeakTflops":312.0,"mfuConfiguredPct":43.893152042386326,"vramAllocatedGb":68.8777856,"vramAllocatedPct":81.05725430142897,"vramReservedGb":71.416414208,"vramReservedPct":84.04478159869937,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.708937216,"vramAllocatedPct":98.510957518641,"vramReservedGb":84.152418304,"vramReservedPct":99.03285814327273,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 498.75 MiB is free. Process 764857 has 78.64 GiB memory in use. Of the allocated memory 77.96 GiB is allocated by PyTorch, and 182.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":26825.575147208885,"meanTps":26819.892466336623,"stepMs":610.7604370117188,"jitter":0.001433529308716091,"achievedTflops":131.6936895688689,"nominalPeakTflops":312.0,"mfuNominalPct":42.20951588745798,"configuredPeakTflops":312.0,"mfuConfiguredPct":42.20951588745798,"vramAllocatedGb":36.901248512,"vramAllocatedPct":43.42639442632446,"vramReservedGb":38.05282304,"vramReservedPct":44.78159869937159,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":27883.37811327591,"meanTps":27882.446388924727,"stepMs":1175.180419921875,"jitter":0.0010330528951775397,"achievedTflops":136.886718037925,"nominalPeakTflops":312.0,"mfuNominalPct":43.87394808907852,"configuredPeakTflops":312.0,"mfuConfiguredPct":43.87394808907852,"vramAllocatedGb":68.8777856,"vramAllocatedPct":81.05725430142897,"vramReservedGb":71.416414208,"vramReservedPct":84.04478159869937,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.708937216,"vramAllocatedPct":98.510957518641,"vramReservedGb":84.152418304,"vramReservedPct":99.03285814327273,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 498.75 MiB is free. Process 852916 has 78.64 GiB memory in use. Of the allocated memory 77.96 GiB is allocated by PyTorch, and 182.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":75352.89452058352,"meanTps":75327.54896835706,"stepMs":434.86053466796875,"jitter":0.00043693984565083285,"achievedTflops":369.9268569137074,"nominalPeakTflops":2250.0,"mfuNominalPct":16.441193640609217,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.441193640609217,"vramAllocatedGb":68.8777856,"vramAllocatedPct":35.96694715733937,"vramReservedGb":71.305265152,"vramReservedPct":37.23454059129997,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":79501.42722345065,"meanTps":79502.23720073464,"stepMs":824.33740234375,"jitter":0.0002454544162185183,"achievedTflops":390.2930773932162,"nominalPeakTflops":2250.0,"mfuNominalPct":17.346358995254054,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.346358995254054,"vramAllocatedGb":132.830859776,"vramAllocatedPct":69.36228383084585,"vramReservedGb":137.705291776,"vramReservedPct":71.90763915315813,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.437265408,"vramAllocatedPct":99.44348532769538,"vramReservedGb":190.559813632,"vramReservedPct":99.50747817326146,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 147.81 MiB is free. Including non-PyTorch memory, this process has 178.19 GiB memory in use. Of the allocated memory 177.25 GiB is allocated by PyTorch, and 128.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":75241.22214678384,"meanTps":75240.31529768658,"stepMs":435.5059509277344,"jitter":0.0010505507278956074,"achievedTflops":369.37862833527475,"nominalPeakTflops":2250.0,"mfuNominalPct":16.41682792601221,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.41682792601221,"vramAllocatedGb":68.8777856,"vramAllocatedPct":35.96694715733937,"vramReservedGb":71.305265152,"vramReservedPct":37.23454059129997,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":79456.18654464076,"meanTps":79411.73645177294,"stepMs":824.8067626953125,"jitter":0.00034158651181466065,"achievedTflops":390.0709791948223,"nominalPeakTflops":2250.0,"mfuNominalPct":17.336487964214324,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.336487964214324,"vramAllocatedGb":132.830859776,"vramAllocatedPct":69.36228383084585,"vramReservedGb":137.705291776,"vramReservedPct":71.90763915315813,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.437265408,"vramAllocatedPct":99.44348532769538,"vramReservedGb":190.559813632,"vramReservedPct":99.50747817326146,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 147.81 MiB is free. Including non-PyTorch memory, this process has 178.19 GiB memory in use. Of the allocated memory 177.25 GiB is allocated by PyTorch, and 128.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":79979.80687234191,"meanTps":79952.11221071574,"stepMs":819.4068298339844,"jitter":0.0003083329165164977,"achievedTflops":392.64156687131424,"nominalPeakTflops":2250.0,"mfuNominalPct":17.450736305391743,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.450736305391743,"vramAllocatedGb":132.830859776,"vramAllocatedPct":46.21350865421075,"vramReservedGb":137.66754304,"vramReservedPct":47.89625093454737,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":16,"tokensPerStep":131072,"status":"complete","stable":true,"tps":82851.15579110455,"meanTps":82850.69998051228,"stepMs":1582.0177612304688,"jitter":9.575206640457852e-05,"achievedTflops":406.7377616807955,"nominalPeakTflops":2250.0,"mfuNominalPct":18.077233852479797,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.077233852479797,"vramAllocatedGb":260.737008128,"vramAllocatedPct":90.71364893606955,"vramReservedGb":270.295629824,"vramReservedPct":94.03921234215868,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.005458944,"vramAllocatedPct":99.50485733773876,"vramReservedGb":286.531780608,"vramReservedPct":99.68797119257023,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 421.62 MiB is free. Including non-PyTorch memory, this process has 267.26 GiB memory in use. Of the allocated memory 266.36 GiB is allocated by PyTorch, and 81.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":80058.91963812719,"meanTps":80056.96953249736,"stepMs":818.5971069335938,"jitter":0.0001463045378747746,"achievedTflops":393.0299519091406,"nominalPeakTflops":2250.0,"mfuNominalPct":17.46799786262847,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.46799786262847,"vramAllocatedGb":132.830859776,"vramAllocatedPct":46.21350865421075,"vramReservedGb":137.66754304,"vramReservedPct":47.89625093454737,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":16,"tokensPerStep":131072,"status":"complete","stable":true,"tps":82890.0044160734,"meanTps":82888.66387088184,"stepMs":1581.2763061523438,"jitter":9.087549533185801e-05,"achievedTflops":406.92847963292695,"nominalPeakTflops":2250.0,"mfuNominalPct":18.085710205907866,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.085710205907866,"vramAllocatedGb":260.737008128,"vramAllocatedPct":90.71364893606955,"vramReservedGb":270.295629824,"vramReservedPct":94.03921234215868,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.005458944,"vramAllocatedPct":99.50485733773876,"vramReservedGb":286.531780608,"vramReservedPct":99.68797119257023,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 421.62 MiB is free. Including non-PyTorch memory, this process has 267.26 GiB memory in use. Of the allocated memory 266.36 GiB is allocated by PyTorch, and 81.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":23467.19884595797,"meanTps":23467.1596613713,"stepMs":349.08299255371094,"jitter":0.000552600963488017,"achievedTflops":115.20655131944346,"nominalPeakTflops":165.2,"mfuNominalPct":69.7376218640699,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":56.05703575676432,"vramAllocatedGb":20.912914432,"vramAllocatedPct":82.78720103358906,"vramReservedGb":20.990394368,"vramReservedPct":83.09391806523753,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.650384384,"vramAllocatedPct":97.58258870082734,"vramReservedGb":24.719130624,"vramReservedPct":97.85473196472722,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 88.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 37.69 MiB is free. Including non-PyTorch memory, this process has 23.48 GiB memory in use. Of the allocated memory 22.96 GiB is allocated by PyTorch, and 65.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":23442.784128379622,"meanTps":23443.203829505706,"stepMs":349.44654846191406,"jitter":0.0007841786736481537,"achievedTflops":115.08669315349437,"nominalPeakTflops":165.2,"mfuNominalPct":69.66506849485131,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":55.99871534514364,"vramAllocatedGb":20.912914432,"vramAllocatedPct":82.78720103358906,"vramReservedGb":20.990394368,"vramReservedPct":83.09391806523753,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.650384384,"vramAllocatedPct":97.58258870082734,"vramReservedGb":24.719130624,"vramReservedPct":97.85473196472722,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 88.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 37.69 MiB is free. Including non-PyTorch memory, this process has 23.48 GiB memory in use. Of the allocated memory 22.96 GiB is allocated by PyTorch, and 65.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":32778.46634195971,"meanTps":32778.880511640535,"stepMs":249.92017364501953,"jitter":0.00041915112848685786,"achievedTflops":160.91797276640315,"nominalPeakTflops":209.5,"mfuNominalPct":76.81048819398718,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":59.83263594916736,"vramAllocatedGb":20.912914432,"vramAllocatedPct":62.11352497148396,"vramReservedGb":20.990394368,"vramReservedPct":62.34364841770058,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.707454976,"vramAllocatedPct":97.14453372872975,"vramReservedGb":32.85188608,"vramReservedPct":97.57350908814861,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 209.88 MiB is free. Including non-PyTorch memory, this process has 31.14 GiB memory in use. Of the allocated memory 30.46 GiB is allocated by PyTorch, and 97.74 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":32763.695565508744,"meanTps":32760.718418445278,"stepMs":250.03284454345703,"jitter":0.0005017051440232843,"achievedTflops":160.84545920283736,"nominalPeakTflops":209.5,"mfuNominalPct":76.77587551448084,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":59.8056739039985,"vramAllocatedGb":20.912914432,"vramAllocatedPct":62.11352497148396,"vramReservedGb":20.990394368,"vramReservedPct":62.34364841770058,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.707454976,"vramAllocatedPct":97.14453372872975,"vramReservedGb":32.85188608,"vramReservedPct":97.57350908814861,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 209.88 MiB is free. Including non-PyTorch memory, this process has 31.14 GiB memory in use. Of the allocated memory 30.46 GiB is allocated by PyTorch, and 97.74 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":54868.93794542867,"meanTps":54735.04704229561,"stepMs":298.6024627685547,"jitter":0.0008060305241739178,"achievedTflops":269.36581382154054,"nominalPeakTflops":989.5,"mfuNominalPct":27.222416758114253,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.222416758114253,"vramAllocatedGb":36.951318016,"vramAllocatedPct":43.463193859345516,"vramReservedGb":38.1157376,"vramReservedPct":44.83281737564597,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":58252.26594224077,"meanTps":58247.619118915536,"stepMs":562.5188903808594,"jitter":0.0003749809617396204,"achievedTflops":285.975446400777,"nominalPeakTflops":989.5,"mfuNominalPct":28.901005194621224,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.901005194621224,"vramAllocatedGb":68.927855104,"vramAllocatedPct":81.07490854309532,"vramReservedGb":71.374471168,"vramReservedPct":83.95268811899506,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.75900672,"vramAllocatedPct":98.51973196088075,"vramReservedGb":84.194361344,"vramReservedPct":99.0318084830844,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 294.19 MiB is free. Including non-PyTorch memory, this process has 78.88 GiB memory in use. Of the allocated memory 78.01 GiB is allocated by PyTorch, and 155.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":54845.92527044674,"meanTps":54833.50040208901,"stepMs":298.7277526855469,"jitter":0.0009538962586595925,"achievedTflops":269.2528386454787,"nominalPeakTflops":989.5,"mfuNominalPct":27.210999357804816,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.210999357804816,"vramAllocatedGb":36.951318016,"vramAllocatedPct":43.463193859345516,"vramReservedGb":38.1157376,"vramReservedPct":44.83281737564597,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":58250.55944019145,"meanTps":58233.55802899064,"stepMs":562.5353698730469,"jitter":0.00040094923501717743,"achievedTflops":285.96706874065615,"nominalPeakTflops":989.5,"mfuNominalPct":28.9001585387222,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.9001585387222,"vramAllocatedGb":68.927855104,"vramAllocatedPct":81.07490854309532,"vramReservedGb":71.374471168,"vramReservedPct":83.95268811899506,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.75900672,"vramAllocatedPct":98.51973196088075,"vramReservedGb":84.194361344,"vramReservedPct":99.0318084830844,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 294.19 MiB is free. Including non-PyTorch memory, this process has 78.88 GiB memory in use. Of the allocated memory 78.01 GiB is allocated by PyTorch, and 155.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":62984.940750590285,"meanTps":62997.72795167314,"stepMs":520.2513427734375,"jitter":0.0015218578267575085,"achievedTflops":309.2093716240368,"nominalPeakTflops":989.5,"mfuNominalPct":31.249052210615137,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.249052210615137,"vramAllocatedGb":68.927855104,"vramAllocatedPct":45.91493903124469,"vramReservedGb":71.389151232,"vramReservedPct":47.55448318192871,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":64467.410567710496,"meanTps":64464.906373579004,"stepMs":1016.5756530761719,"jitter":0.0015773945067223886,"achievedTflops":316.48719954831006,"nominalPeakTflops":989.5,"mfuNominalPct":31.98455781185549,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.98455781185549,"vramAllocatedGb":132.88092928,"vramAllocatedPct":88.51602530066648,"vramReservedGb":137.789177856,"vramReservedPct":91.78555884087606,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.123220992,"vramAllocatedPct":98.66937902967746,"vramReservedGb":149.04459264,"vramReservedPct":99.28313268528167,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.03 GiB is free. Including non-PyTorch memory, this process has 138.77 GiB memory in use. Of the allocated memory 137.95 GiB is allocated by PyTorch, and 98.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":62848.33602093106,"meanTps":62877.3569906796,"stepMs":521.3821411132812,"jitter":0.0012706276804031688,"achievedTflops":308.5387436593927,"nominalPeakTflops":989.5,"mfuNominalPct":31.181277782657173,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.181277782657173,"vramAllocatedGb":68.927855104,"vramAllocatedPct":45.91493903124469,"vramReservedGb":71.389151232,"vramReservedPct":47.55448318192871,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":64399.784430763495,"meanTps":64389.548297687645,"stepMs":1017.6431579589844,"jitter":0.0008357765643036668,"achievedTflops":316.155205343639,"nominalPeakTflops":989.5,"mfuNominalPct":31.951006098397073,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.951006098397073,"vramAllocatedGb":132.88092928,"vramAllocatedPct":88.51602530066648,"vramReservedGb":137.789177856,"vramReservedPct":91.78555884087606,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.123220992,"vramAllocatedPct":98.66937902967746,"vramReservedGb":149.04459264,"vramReservedPct":99.28313268528167,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.03 GiB is free. Including non-PyTorch memory, this process has 138.77 GiB memory in use. Of the allocated memory 137.95 GiB is allocated by PyTorch, and 98.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":22617.920868629146,"meanTps":22612.863477407373,"stepMs":362.190673828125,"jitter":0.005184987088254862,"achievedTflops":111.03722597636107,"nominalPeakTflops":364.2,"mfuNominalPct":30.48798077330068,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":24.596583522421323,"vramAllocatedGb":20.912914432,"vramAllocatedPct":41.104977464160996,"vramReservedGb":20.990394368,"vramReservedPct":41.25726666486329,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21429.420680164116,"meanTps":21428.127875419716,"stepMs":764.5563659667969,"jitter":0.0023496855843473806,"achievedTflops":105.20257102438546,"nominalPeakTflops":364.2,"mfuNominalPct":28.885933834262897,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":23.304110871120308,"vramAllocatedGb":36.901248512,"vramAllocatedPct":72.53054056225598,"vramReservedGb":38.075891712,"vramReservedPct":74.83933795256847,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14275.272043886729,"meanTps":14274.494790175866,"stepMs":573.8594665527344,"jitter":0.0009988743169288478,"achievedTflops":70.08100421863185,"nominalPeakTflops":154.8,"mfuNominalPct":45.27196654950377,"configuredPeakTflops":180.6,"mfuConfiguredPct":38.804542756717524,"vramAllocatedGb":20.912914432,"vramAllocatedPct":40.978450423517685,"vramReservedGb":20.990394368,"vramReservedPct":41.13027085612727,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":15175.787255876789,"meanTps":15174.806509462265,"stepMs":1079.614501953125,"jitter":0.0006229325834032285,"achievedTflops":74.50186640440319,"nominalPeakTflops":154.8,"mfuNominalPct":48.12782067467906,"configuredPeakTflops":180.6,"mfuConfiguredPct":41.252417721153485,"vramAllocatedGb":36.901248512,"vramAllocatedPct":72.3072811124338,"vramReservedGb":38.075891712,"vramReservedPct":74.60897169186198,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.493198336,"vramAllocatedPct":98.94044330667407,"vramReservedGb":50.629443584,"vramReservedPct":99.20741322895638,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 64.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 48.25 MiB is free. Process 1415479 has 47.47 GiB memory in use. Of the allocated memory 47.03 GiB is allocated by PyTorch, and 129.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14302.962592036518,"meanTps":14307.17001326724,"stepMs":572.7484741210938,"jitter":0.0009415642049461039,"achievedTflops":70.21694428448389,"nominalPeakTflops":154.8,"mfuNominalPct":45.359783129511555,"configuredPeakTflops":180.6,"mfuConfiguredPct":38.87981411100991,"vramAllocatedGb":20.912914432,"vramAllocatedPct":40.978450423517685,"vramReservedGb":20.990394368,"vramReservedPct":41.13027085612727,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":15185.971102266494,"meanTps":15187.612939768795,"stepMs":1078.8905029296875,"jitter":0.0006813913264257228,"achievedTflops":74.55186154142088,"nominalPeakTflops":154.8,"mfuNominalPct":48.16011727481968,"configuredPeakTflops":180.6,"mfuConfiguredPct":41.28010052127402,"vramAllocatedGb":36.901248512,"vramAllocatedPct":72.3072811124338,"vramReservedGb":38.075891712,"vramReservedPct":74.60897169186198,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.493198336,"vramAllocatedPct":98.94044330667407,"vramReservedGb":50.629443584,"vramReservedPct":99.20741322895638,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 64.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 48.25 MiB is free. Process 1449495 has 47.47 GiB memory in use. Of the allocated memory 47.03 GiB is allocated by PyTorch, and 129.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":44077.51790966829,"meanTps":44072.68263988049,"stepMs":185.85438537597656,"jitter":0.0003065583905226806,"achievedTflops":216.38794056447583,"nominalPeakTflops":468.0,"mfuNominalPct":46.23673943685381,"configuredPeakTflops":468.0,"mfuConfiguredPct":46.23673943685381,"vramAllocatedGb":20.912914432,"vramAllocatedPct":20.508094799546786,"vramReservedGb":20.990394368,"vramReservedPct":20.584074925498022,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":44980.1488472008,"meanTps":44979.362934136414,"stepMs":364.24957275390625,"jitter":0.00030637697903689173,"achievedTflops":220.81918939438324,"nominalPeakTflops":468.0,"mfuNominalPct":47.18358747743232,"configuredPeakTflops":468.0,"mfuConfiguredPct":47.18358747743232,"vramAllocatedGb":36.901248512,"vramAllocatedPct":36.186936314708426,"vramReservedGb":38.075891712,"vramReservedPct":37.33884147740455,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":41975.48081593528,"meanTps":41975.90894035263,"stepMs":780.6462097167969,"jitter":0.0002112020716010401,"achievedTflops":206.0684965650384,"nominalPeakTflops":468.0,"mfuNominalPct":44.031730035264616,"configuredPeakTflops":468.0,"mfuConfiguredPct":44.031730035264616,"vramAllocatedGb":68.8777856,"vramAllocatedPct":67.54449081024474,"vramReservedGb":71.424802816,"vramReservedPct":70.04220439730359,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":43933.74850751981,"meanTps":43932.39296797972,"stepMs":186.46257781982422,"jitter":0.0006567353721393797,"achievedTflops":215.6821393687086,"nominalPeakTflops":468.0,"mfuNominalPct":46.085927215536024,"configuredPeakTflops":468.0,"mfuConfiguredPct":46.085927215536024,"vramAllocatedGb":21.032772096,"vramAllocatedPct":20.62563233090124,"vramReservedGb":22.204645376,"vramReservedPct":21.774821192044467,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":44885.81917943959,"meanTps":44888.011449379104,"stepMs":365.0150604248047,"jitter":0.0003620099335216622,"achievedTflops":220.3561006473533,"nominalPeakTflops":468.0,"mfuNominalPct":47.08463689046011,"configuredPeakTflops":468.0,"mfuConfiguredPct":47.08463689046011,"vramAllocatedGb":37.00464896,"vramAllocatedPct":36.28833519896221,"vramReservedGb":40.995127296,"vramReservedPct":40.20156825293589,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":41953.49395851412,"meanTps":41949.94027357414,"stepMs":781.0553283691406,"jitter":0.0001153935662327734,"achievedTflops":205.96055739281456,"nominalPeakTflops":468.0,"mfuNominalPct":44.008666109575756,"configuredPeakTflops":468.0,"mfuConfiguredPct":44.008666109575756,"vramAllocatedGb":69.010661888,"vramAllocatedPct":67.67479495889779,"vramReservedGb":71.793901568,"vramReservedPct":70.40415835742824,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":20899.160737812657,"meanTps":20899.495196318345,"stepMs":783.9549255371094,"jitter":0.0009842009044077273,"achievedTflops":161.50518305961793,"nominalPeakTflops":312.0,"mfuNominalPct":51.76448174987755,"configuredPeakTflops":312.0,"mfuConfiguredPct":51.76448174987755,"vramAllocatedGb":36.930797568,"vramAllocatedPct":43.46116856033145,"vramReservedGb":37.033607168,"vramReservedPct":43.58215769810983,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":21452.133258330046,"meanTps":21453.4329048732,"stepMs":1527.4937744140625,"jitter":0.0011523618404170497,"achievedTflops":165.77846126793926,"nominalPeakTflops":312.0,"mfuNominalPct":53.134122201262585,"configuredPeakTflops":312.0,"mfuConfiguredPct":53.134122201262585,"vramAllocatedGb":68.907785728,"vramAllocatedPct":81.0925592692526,"vramReservedGb":71.443677184,"vramReservedPct":84.07686541149032,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.738297344,"vramAllocatedPct":98.54550931703126,"vramReservedGb":84.194361344,"vramReservedPct":99.08221785525882,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 458.75 MiB is free. Process 925444 has 78.68 GiB memory in use. Of the allocated memory 77.99 GiB is allocated by PyTorch, and 194.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":20878.65403047523,"meanTps":20877.616314815827,"stepMs":784.7249145507812,"jitter":0.0009484099901586171,"achievedTflops":161.3467106901276,"nominalPeakTflops":312.0,"mfuNominalPct":51.71368932375885,"configuredPeakTflops":312.0,"mfuConfiguredPct":51.71368932375885,"vramAllocatedGb":36.930797568,"vramAllocatedPct":43.46116856033145,"vramReservedGb":37.033607168,"vramReservedPct":43.58215769810983,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":21468.005360694224,"meanTps":21469.507919890348,"stepMs":1526.3644409179688,"jitter":0.0006772141024725932,"achievedTflops":165.9011181932592,"nominalPeakTflops":312.0,"mfuNominalPct":53.17343531835231,"configuredPeakTflops":312.0,"mfuConfiguredPct":53.17343531835231,"vramAllocatedGb":68.907785728,"vramAllocatedPct":81.0925592692526,"vramReservedGb":71.443677184,"vramReservedPct":84.07686541149032,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.738297344,"vramAllocatedPct":98.54550931703126,"vramReservedGb":84.194361344,"vramReservedPct":99.08221785525882,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 458.75 MiB is free. Process 937161 has 78.68 GiB memory in use. Of the allocated memory 77.99 GiB is allocated by PyTorch, and 194.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":53734.15156994496,"meanTps":53673.74817457255,"stepMs":609.8170166015625,"jitter":0.0011983973163696767,"achievedTflops":415.2484444102854,"nominalPeakTflops":2250.0,"mfuNominalPct":18.455486418234905,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.455486418234905,"vramAllocatedGb":68.907785728,"vramAllocatedPct":35.982612774476884,"vramReservedGb":71.33462528,"vramReservedPct":37.249872010031126,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":55918.61583042941,"meanTps":55895.736611390734,"stepMs":1171.9889526367188,"jitter":0.0012985386681409678,"achievedTflops":432.1296151282277,"nominalPeakTflops":2250.0,"mfuNominalPct":19.205760672365678,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.205760672365678,"vramAllocatedGb":132.861499904,"vramAllocatedPct":69.37828364638972,"vramReservedGb":137.732554752,"vramReservedPct":71.92187547055136,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.466691072,"vramAllocatedPct":99.45885096834336,"vramReservedGb":190.58917376,"vramReservedPct":99.52280959199263,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 119.81 MiB is free. Including non-PyTorch memory, this process has 178.22 GiB memory in use. Of the allocated memory 177.28 GiB is allocated by PyTorch, and 128.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":53705.74320461496,"meanTps":53708.06742281265,"stepMs":610.1395874023438,"jitter":0.0002566530656096862,"achievedTflops":415.02890936289253,"nominalPeakTflops":2250.0,"mfuNominalPct":18.445729305017444,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.445729305017444,"vramAllocatedGb":68.907785728,"vramAllocatedPct":35.982612774476884,"vramReservedGb":71.33462528,"vramReservedPct":37.249872010031126,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":56000.01794100382,"meanTps":55997.59450755348,"stepMs":1170.2853393554688,"jitter":0.0002642148789855822,"achievedTflops":432.7586768850462,"nominalPeakTflops":2250.0,"mfuNominalPct":19.233718972668722,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.233718972668722,"vramAllocatedGb":132.861499904,"vramAllocatedPct":69.37828364638972,"vramReservedGb":137.732554752,"vramReservedPct":71.92187547055136,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.466691072,"vramAllocatedPct":99.45885096834336,"vramReservedGb":190.58917376,"vramReservedPct":99.52280959199263,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 119.81 MiB is free. Including non-PyTorch memory, this process has 178.22 GiB memory in use. Of the allocated memory 177.28 GiB is allocated by PyTorch, and 128.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":56507.5229386767,"meanTps":56497.98576767148,"stepMs":1159.7747802734375,"jitter":0.00016509961450918994,"achievedTflops":436.68058975901783,"nominalPeakTflops":2250.0,"mfuNominalPct":19.408026211511903,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.408026211511903,"vramAllocatedGb":132.861499904,"vramAllocatedPct":46.22416873593334,"vramReservedGb":137.696903168,"vramReservedPct":47.906465688345584,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":8,"tokensPerStep":131072,"status":"complete","stable":true,"tps":57998.02651265947,"meanTps":57990.64511283121,"stepMs":2259.93896484375,"jitter":0.00013453851314343865,"achievedTflops":448.1989495432728,"nominalPeakTflops":2250.0,"mfuNominalPct":19.91995331303435,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.91995331303435,"vramAllocatedGb":260.768928256,"vramAllocatedPct":90.72475434571652,"vramReservedGb":270.327087104,"vramReservedPct":94.05015672122819,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.034819072,"vramAllocatedPct":99.51507209153696,"vramReservedGb":286.120738816,"vramReservedPct":99.54496463939525,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 393.62 MiB is free. Including non-PyTorch memory, this process has 267.29 GiB memory in use. Of the allocated memory 266.39 GiB is allocated by PyTorch, and 81.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":56511.644934802294,"meanTps":56508.43055892255,"stepMs":1159.690185546875,"jitter":0.0001271890370100278,"achievedTflops":436.712443848624,"nominalPeakTflops":2250.0,"mfuNominalPct":19.409441948827734,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.409441948827734,"vramAllocatedGb":132.861499904,"vramAllocatedPct":46.22416873593334,"vramReservedGb":137.696903168,"vramReservedPct":47.906465688345584,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":8,"tokensPerStep":131072,"status":"complete","stable":true,"tps":57988.75190567746,"meanTps":57986.221558790596,"stepMs":2260.3004150390625,"jitter":0.00013739950587536516,"achievedTflops":448.1272769475534,"nominalPeakTflops":2250.0,"mfuNominalPct":19.916767864335707,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.916767864335707,"vramAllocatedGb":260.768928256,"vramAllocatedPct":90.72475434571652,"vramReservedGb":270.327087104,"vramReservedPct":94.05015672122819,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.034819072,"vramAllocatedPct":99.51507209153696,"vramReservedGb":286.120738816,"vramReservedPct":99.54496463939525,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 393.62 MiB is free. Including non-PyTorch memory, this process has 267.29 GiB memory in use. Of the allocated memory 266.39 GiB is allocated by PyTorch, and 81.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.687422464,"vramAllocatedPct":97.72921001393166,"vramReservedGb":24.72542208,"vramReservedPct":97.87963772496258,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 88.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 31.69 MiB is free. Including non-PyTorch memory, this process has 23.49 GiB memory in use. Of the allocated memory 22.99 GiB is allocated by PyTorch, and 36.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.687422464,"vramAllocatedPct":97.72921001393166,"vramReservedGb":24.72542208,"vramReservedPct":97.87963772496258,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 88.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 31.69 MiB is free. Including non-PyTorch memory, this process has 23.49 GiB memory in use. Of the allocated memory 22.99 GiB is allocated by PyTorch, and 36.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.046075904,"vramAllocatedPct":95.18016928598958,"vramReservedGb":32.155631616,"vramReservedPct":95.50556111385782,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 873.88 MiB is free. Including non-PyTorch memory, this process has 30.49 GiB memory in use. Of the allocated memory 29.85 GiB is allocated by PyTorch, and 64.48 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.046075904,"vramAllocatedPct":95.18016928598958,"vramReservedGb":32.155631616,"vramReservedPct":95.50556111385782,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 31.36 GiB of which 873.88 MiB is free. Including non-PyTorch memory, this process has 30.49 GiB memory in use. Of the allocated memory 29.85 GiB is allocated by PyTorch, and 64.48 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":42438.49953555427,"meanTps":42435.38763380667,"stepMs":386.0645446777344,"jitter":0.0004526709601778197,"achievedTflops":327.9575540018819,"nominalPeakTflops":989.5,"mfuNominalPct":33.14376493197391,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.14376493197391,"vramAllocatedGb":36.980867072,"vramAllocatedPct":43.49795030155774,"vramReservedGb":37.096521728,"vramReservedPct":43.63398660565621,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":44097.07895439331,"meanTps":44091.532542180474,"stepMs":743.0877685546875,"jitter":0.000355039758516602,"achievedTflops":340.77477551709137,"nominalPeakTflops":989.5,"mfuNominalPct":34.43908797545138,"configuredPeakTflops":989.5,"mfuConfiguredPct":34.43908797545138,"vramAllocatedGb":68.957855232,"vramAllocatedPct":81.11019554905555,"vramReservedGb":71.401734144,"vramReservedPct":83.9847556087273,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.788366848,"vramAllocatedPct":98.55426618059238,"vramReservedGb":84.110475264,"vramReservedPct":98.93313928390829,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 254.19 MiB is free. Including non-PyTorch memory, this process has 78.92 GiB memory in use. Of the allocated memory 78.03 GiB is allocated by PyTorch, and 167.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":42420.72553958883,"meanTps":42408.582198991724,"stepMs":386.22630310058594,"jitter":0.000876847399626268,"achievedTflops":327.82019956415525,"nominalPeakTflops":989.5,"mfuNominalPct":33.129883735639744,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.129883735639744,"vramAllocatedGb":36.980867072,"vramAllocatedPct":43.49795030155774,"vramReservedGb":37.096521728,"vramReservedPct":43.63398660565621,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":44082.66625162614,"meanTps":44083.71669728512,"stepMs":743.3307189941406,"jitter":0.00038462750284194236,"achievedTflops":340.663396585277,"nominalPeakTflops":989.5,"mfuNominalPct":34.42783189340849,"configuredPeakTflops":989.5,"mfuConfiguredPct":34.42783189340849,"vramAllocatedGb":68.957855232,"vramAllocatedPct":81.11019554905555,"vramReservedGb":71.401734144,"vramReservedPct":83.9847556087273,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.788366848,"vramAllocatedPct":98.55426618059238,"vramReservedGb":84.110475264,"vramReservedPct":98.93313928390829,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 254.19 MiB is free. Including non-PyTorch memory, this process has 78.92 GiB memory in use. Of the allocated memory 78.03 GiB is allocated by PyTorch, and 167.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":46766.08681794328,"meanTps":46743.59191735564,"stepMs":700.6786804199219,"jitter":0.0018999497252572197,"achievedTflops":361.40041733103686,"nominalPeakTflops":989.5,"mfuNominalPct":36.523538891464064,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.523538891464064,"vramAllocatedGb":68.957855232,"vramAllocatedPct":45.93492302822199,"vramReservedGb":71.41851136,"vramReservedPct":47.57404085545613,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":47429.3902523921,"meanTps":47439.04454973516,"stepMs":1381.75927734375,"jitter":0.0008505776115494746,"achievedTflops":366.52631420070907,"nominalPeakTflops":989.5,"mfuNominalPct":37.04156788284074,"configuredPeakTflops":989.5,"mfuConfiguredPct":37.04156788284074,"vramAllocatedGb":132.911569408,"vramAllocatedPct":88.53643562109364,"vramReservedGb":137.837412352,"vramReservedPct":91.81768930452826,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.15258112,"vramAllocatedPct":98.68893670320489,"vramReservedGb":149.073952768,"vramReservedPct":99.3026903588091,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.01 GiB is free. Including non-PyTorch memory, this process has 138.80 GiB memory in use. Of the allocated memory 137.98 GiB is allocated by PyTorch, and 98.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":46739.75480714693,"meanTps":46739.61523594821,"stepMs":701.0734252929688,"jitter":0.002316694196356086,"achievedTflops":361.1969280006592,"nominalPeakTflops":989.5,"mfuNominalPct":36.50297402735313,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.50297402735313,"vramAllocatedGb":68.957855232,"vramAllocatedPct":45.93492302822199,"vramReservedGb":71.41851136,"vramReservedPct":47.57404085545613,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":47472.871020610175,"meanTps":47454.24201748715,"stepMs":1380.4937133789062,"jitter":0.0015925915318187377,"achievedTflops":366.8623262309878,"nominalPeakTflops":989.5,"mfuNominalPct":37.07552564234338,"configuredPeakTflops":989.5,"mfuConfiguredPct":37.07552564234338,"vramAllocatedGb":132.911569408,"vramAllocatedPct":88.53643562109364,"vramReservedGb":137.837412352,"vramReservedPct":91.81768930452826,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.15258112,"vramAllocatedPct":98.68893670320489,"vramReservedGb":149.073952768,"vramReservedPct":99.3026903588091,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.01 GiB is free. Including non-PyTorch memory, this process has 138.80 GiB memory in use. Of the allocated memory 137.98 GiB is allocated by PyTorch, and 98.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":12283.277003521549,"meanTps":12283.091242962664,"stepMs":1333.8460083007812,"jitter":0.00064728010541723,"achievedTflops":94.92308930073202,"nominalPeakTflops":154.8,"mfuNominalPct":61.319825129671834,"configuredPeakTflops":180.6,"mfuConfiguredPct":52.55985011114729,"vramAllocatedGb":36.930797568,"vramAllocatedPct":72.36518191484444,"vramReservedGb":37.033607168,"vramReservedPct":72.5666353330354,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.522558464,"vramAllocatedPct":98.99797390833115,"vramReservedGb":50.681872384,"vramReservedPct":99.31014644620119,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 64.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 256.00 KiB is free. Process 1472317 has 47.52 GiB memory in use. Of the allocated memory 47.05 GiB is allocated by PyTorch, and 151.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":12286.715009389125,"meanTps":12285.576852929673,"stepMs":1333.4727783203125,"jitter":0.00047726602506506657,"achievedTflops":94.94965762919117,"nominalPeakTflops":154.8,"mfuNominalPct":61.33698813255243,"configuredPeakTflops":180.6,"mfuConfiguredPct":52.574561256473515,"vramAllocatedGb":36.930797568,"vramAllocatedPct":72.36518191484444,"vramReservedGb":37.033607168,"vramReservedPct":72.5666353330354,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.522558464,"vramAllocatedPct":98.99797390833115,"vramReservedGb":50.681872384,"vramReservedPct":99.31014644620119,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 64.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 256.00 KiB is free. Process 1479887 has 47.52 GiB memory in use. Of the allocated memory 47.05 GiB is allocated by PyTorch, and 151.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":36953.31601708122,"meanTps":36948.32602970771,"stepMs":443.37022399902344,"jitter":0.0002736436274566657,"achievedTflops":285.56898254772943,"nominalPeakTflops":468.0,"mfuNominalPct":61.01901336489945,"configuredPeakTflops":468.0,"mfuConfiguredPct":61.01901336489945,"vramAllocatedGb":36.972012032,"vramAllocatedPct":36.25633003700516,"vramReservedGb":39.787167744,"vramReservedPct":39.0169916561643,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":35061.77381370544,"meanTps":35061.95127424346,"stepMs":934.5790710449219,"jitter":7.673390077620418e-05,"achievedTflops":270.9514639950121,"nominalPeakTflops":468.0,"mfuNominalPct":57.89561196474617,"configuredPeakTflops":468.0,"mfuConfiguredPct":57.89561196474617,"vramAllocatedGb":68.98130176,"vramAllocatedPct":67.64600316661515,"vramReservedGb":71.80648448,"vramReservedPct":70.41649769697796,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":36949.27224606589,"meanTps":36951.12667570377,"stepMs":443.4187469482422,"jitter":0.00022116638359035828,"achievedTflops":285.5377329685585,"nominalPeakTflops":468.0,"mfuNominalPct":61.01233610439284,"configuredPeakTflops":468.0,"mfuConfiguredPct":61.01233610439284,"vramAllocatedGb":36.972012032,"vramAllocatedPct":36.25633003700516,"vramReservedGb":39.787167744,"vramReservedPct":39.0169916561643,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":35061.02506450543,"meanTps":35061.21888902114,"stepMs":934.5990295410156,"jitter":0.00019911658931849043,"achievedTflops":270.94567778770306,"nominalPeakTflops":468.0,"mfuNominalPct":57.89437559566305,"configuredPeakTflops":468.0,"mfuConfiguredPct":57.89437559566305,"vramAllocatedGb":68.98130176,"vramAllocatedPct":67.64600316661515,"vramReservedGb":71.80648448,"vramReservedPct":70.41649769697796,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":14737.219196019589,"meanTps":14734.290586623623,"stepMs":2223.4859619140625,"jitter":0.0005082326418460771,"achievedTflops":196.96257520244893,"nominalPeakTflops":312.0,"mfuNominalPct":63.12903051360543,"configuredPeakTflops":312.0,"mfuConfiguredPct":63.12903051360543,"vramAllocatedGb":68.96656384,"vramAllocatedPct":81.16173095254989,"vramReservedGb":69.367496704,"vramReservedPct":81.63355966817933,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.7970176,"vramAllocatedPct":98.61461291381177,"vramReservedGb":84.246790144,"vramReservedPct":99.14391749524141,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 408.75 MiB is free. Process 949988 has 78.73 GiB memory in use. Of the allocated memory 78.04 GiB is allocated by PyTorch, and 188.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":14752.351559918317,"meanTps":14749.115139759242,"stepMs":2221.2052001953125,"jitter":0.0006254323437098901,"achievedTflops":197.16481887696787,"nominalPeakTflops":312.0,"mfuNominalPct":63.19385220415637,"configuredPeakTflops":312.0,"mfuConfiguredPct":63.19385220415637,"vramAllocatedGb":68.96656384,"vramAllocatedPct":81.16173095254989,"vramReservedGb":69.367496704,"vramReservedPct":81.63355966817933,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.7970176,"vramAllocatedPct":98.61461291381177,"vramReservedGb":84.246790144,"vramReservedPct":99.14391749524141,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 408.75 MiB is free. Process 970883 has 78.73 GiB memory in use. Of the allocated memory 78.04 GiB is allocated by PyTorch, and 188.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":34540.92736187915,"meanTps":34537.88244470408,"stepMs":948.6716918945312,"jitter":0.00019611518649797975,"achievedTflops":461.638651945542,"nominalPeakTflops":2250.0,"mfuNominalPct":20.517273419801867,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.517273419801867,"vramAllocatedGb":68.96656384,"vramAllocatedPct":36.01330582347514,"vramReservedGb":69.409439744,"vramReservedPct":36.24456898180215,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":35463.1355298241,"meanTps":35456.825653872926,"stepMs":1848.0035400390625,"jitter":0.0006580641191583637,"achievedTflops":473.9639416230024,"nominalPeakTflops":2250.0,"mfuNominalPct":21.06506407213344,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.06506407213344,"vramAllocatedGb":132.92086016,"vramAllocatedPct":69.4092806822584,"vramReservedGb":137.847898112,"vramReservedPct":71.98210604413806,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.391324672,"vramAllocatedPct":99.41949576401113,"vramReservedGb":190.490607616,"vramReservedPct":99.47133982910944,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 115.81 MiB is free. Including non-PyTorch memory, this process has 178.22 GiB memory in use. Of the allocated memory 177.32 GiB is allocated by PyTorch, and 92.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":34429.70062141313,"meanTps":34416.59403067317,"stepMs":951.7364196777344,"jitter":0.0009091511810928183,"achievedTflops":460.15210927136656,"nominalPeakTflops":2250.0,"mfuNominalPct":20.451204856505182,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.451204856505182,"vramAllocatedGb":68.96656384,"vramAllocatedPct":36.01330582347514,"vramReservedGb":69.409439744,"vramReservedPct":36.24456898180215,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":35409.068259705986,"meanTps":33953.73524778815,"stepMs":1850.8253173828125,"jitter":0.0007563117791953164,"achievedTflops":473.24133387625005,"nominalPeakTflops":2250.0,"mfuNominalPct":21.03294817227778,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.03294817227778,"vramAllocatedGb":132.92086016,"vramAllocatedPct":69.4092806822584,"vramReservedGb":137.847898112,"vramReservedPct":71.98210604413806,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.391324672,"vramAllocatedPct":99.41949576401113,"vramReservedGb":190.490607616,"vramReservedPct":99.47133982910944,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 115.75 MiB is free. Including non-PyTorch memory, this process has 178.22 GiB memory in use. Of the allocated memory 177.32 GiB is allocated by PyTorch, and 92.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":35843.23420296107,"meanTps":35839.59253148807,"stepMs":1828.4064331054688,"jitter":0.00010259604465939087,"achievedTflops":479.04395112115196,"nominalPeakTflops":2250.0,"mfuNominalPct":21.2908422720512,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.2908422720512,"vramAllocatedGb":132.92086016,"vramAllocatedPct":46.24482090749195,"vramReservedGb":137.910812672,"vramReservedPct":47.980887466018274,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":4,"tokensPerStep":131072,"status":"complete","stable":true,"tps":36486.32731796217,"meanTps":36485.63984788898,"stepMs":3592.359375,"jitter":7.340300295549147e-05,"achievedTflops":487.6388749219591,"nominalPeakTflops":2250.0,"mfuNominalPct":21.672838885420404,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.672838885420404,"vramAllocatedGb":260.828928512,"vramAllocatedPct":90.74562918123732,"vramReservedGb":270.840889344,"vramReservedPct":94.22891491269692,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.093539328,"vramAllocatedPct":99.53550159913338,"vramReservedGb":286.25076224,"vramReservedPct":99.59020140621591,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 269.62 MiB is free. Including non-PyTorch memory, this process has 267.41 GiB memory in use. Of the allocated memory 266.45 GiB is allocated by PyTorch, and 149.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":35855.3529118575,"meanTps":35853.650052748424,"stepMs":1827.7884521484375,"jitter":0.00010758156399385222,"achievedTflops":479.20591737005014,"nominalPeakTflops":2250.0,"mfuNominalPct":21.298040772002228,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.298040772002228,"vramAllocatedGb":132.92086016,"vramAllocatedPct":46.24482090749195,"vramReservedGb":137.910812672,"vramReservedPct":47.980887466018274,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":4,"tokensPerStep":131072,"status":"complete","stable":true,"tps":36485.79296116329,"meanTps":36485.12956223855,"stepMs":3592.4119873046875,"jitter":9.012769326242696e-05,"achievedTflops":487.63173325637183,"nominalPeakTflops":2250.0,"mfuNominalPct":21.67252147806097,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.67252147806097,"vramAllocatedGb":260.828928512,"vramAllocatedPct":90.74562918123732,"vramReservedGb":270.840889344,"vramReservedPct":94.22891491269692,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.093539328,"vramAllocatedPct":99.53550159913338,"vramReservedGb":286.25076224,"vramReservedPct":99.59020140621591,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 269.62 MiB is free. Including non-PyTorch memory, this process has 267.41 GiB memory in use. Of the allocated memory 266.45 GiB is allocated by PyTorch, and 149.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.795266048,"vramAllocatedPct":97.40534172042216,"vramReservedGb":32.9252864,"vramReservedPct":97.79151565170338,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 176.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 141.88 MiB is free. Including non-PyTorch memory, this process has 31.21 GiB memory in use. Of the allocated memory 30.54 GiB is allocated by PyTorch, and 82.00 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.795266048,"vramAllocatedPct":97.40534172042216,"vramReservedGb":32.9252864,"vramReservedPct":97.79151565170338,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 176.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 141.88 MiB is free. Including non-PyTorch memory, this process has 31.21 GiB memory in use. Of the allocated memory 30.54 GiB is allocated by PyTorch, and 82.00 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_32k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":29650.47611778521,"meanTps":29650.353138601935,"stepMs":1105.1424560546875,"jitter":0.00029248072007119594,"achievedTflops":396.27789031698967,"nominalPeakTflops":989.5,"mfuNominalPct":40.04829614118137,"configuredPeakTflops":989.5,"mfuConfiguredPct":40.04829614118137,"vramAllocatedGb":69.016633344,"vramAllocatedPct":81.1793320403557,"vramReservedGb":69.451382784,"vramReservedPct":81.69069672788268,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_32k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.847087104,"vramAllocatedPct":98.62333462001567,"vramReservedGb":84.204847104,"vramReservedPct":99.0441421329814,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 224.19 MiB is free. Including non-PyTorch memory, this process has 78.95 GiB memory in use. Of the allocated memory 78.09 GiB is allocated by PyTorch, and 141.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_32k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":29649.81456402249,"meanTps":29645.410252835045,"stepMs":1105.1671142578125,"jitter":0.00040104664159136175,"achievedTflops":396.26904866708225,"nominalPeakTflops":989.5,"mfuNominalPct":40.047402593944646,"configuredPeakTflops":989.5,"mfuConfiguredPct":40.047402593944646,"vramAllocatedGb":69.016633344,"vramAllocatedPct":81.1793320403557,"vramReservedGb":69.451382784,"vramReservedPct":81.69069672788268,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_32k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.847087104,"vramAllocatedPct":98.62333462001567,"vramReservedGb":84.204847104,"vramReservedPct":99.0441421329814,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 704.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 224.19 MiB is free. Including non-PyTorch memory, this process has 78.95 GiB memory in use. Of the allocated memory 78.09 GiB is allocated by PyTorch, and 141.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30867.037731291817,"meanTps":30868.925041290357,"stepMs":1061.5855102539062,"jitter":0.0009937408098893942,"achievedTflops":412.53720661686714,"nominalPeakTflops":989.5,"mfuNominalPct":41.69148121443831,"configuredPeakTflops":989.5,"mfuConfiguredPct":41.69148121443831,"vramAllocatedGb":69.016633344,"vramAllocatedPct":45.9740769149167,"vramReservedGb":69.451382784,"vramReservedPct":46.2636767291188,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":31204.58260839341,"meanTps":31199.86659750757,"stepMs":2100.204345703125,"jitter":0.0013639502852505632,"achievedTflops":417.04848566863564,"nominalPeakTflops":989.5,"mfuNominalPct":42.147396227249686,"configuredPeakTflops":989.5,"mfuConfiguredPct":42.147396227249686,"vramAllocatedGb":132.970929664,"vramAllocatedPct":88.57597729159836,"vramReservedGb":138.015670272,"vramReservedPct":91.93643232237332,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.211301376,"vramAllocatedPct":98.72805205025973,"vramReservedGb":149.080244224,"vramReservedPct":99.30688128885069,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 943.44 MiB is free. Including non-PyTorch memory, this process has 138.88 GiB memory in use. Of the allocated memory 138.03 GiB is allocated by PyTorch, and 128.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30856.90054061989,"meanTps":30844.3835441532,"stepMs":1061.9342651367188,"jitter":0.0010501663828241265,"achievedTflops":412.401723310722,"nominalPeakTflops":989.5,"mfuNominalPct":41.677789116798586,"configuredPeakTflops":989.5,"mfuConfiguredPct":41.677789116798586,"vramAllocatedGb":69.016633344,"vramAllocatedPct":45.9740769149167,"vramReservedGb":69.451382784,"vramReservedPct":46.2636767291188,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":31151.506064573514,"meanTps":31150.70715182165,"stepMs":2103.78271484375,"jitter":0.0009372606936862937,"achievedTflops":416.3391189547011,"nominalPeakTflops":989.5,"mfuNominalPct":42.07570681704913,"configuredPeakTflops":989.5,"mfuConfiguredPct":42.07570681704913,"vramAllocatedGb":132.970929664,"vramAllocatedPct":88.57597729159836,"vramReservedGb":138.015670272,"vramReservedPct":91.93643232237332,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.211301376,"vramAllocatedPct":98.72805205025973,"vramReservedGb":149.080244224,"vramReservedPct":99.30688128885069,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.38 GiB. GPU 0 has a total capacity of 139.81 GiB of which 943.44 MiB is free. Including non-PyTorch memory, this process has 138.88 GiB memory in use. Of the allocated memory 138.03 GiB is allocated by PyTorch, and 128.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.523026432,"vramAllocatedPct":98.99889088255539,"vramReservedGb":50.679775232,"vramReservedPct":99.3060371175114,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 62.25 MiB is free. Process 1490581 has 47.46 GiB memory in use. Of the allocated memory 47.05 GiB is allocated by PyTorch, and 89.49 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.523026432,"vramAllocatedPct":98.99889088255539,"vramReservedGb":50.679775232,"vramReservedPct":99.3060371175114,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 62.25 MiB is free. Process 1493539 has 47.46 GiB memory in use. Of the allocated memory 47.05 GiB is allocated by PyTorch, and 89.49 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":26523.752692351227,"meanTps":26522.04759344629,"stepMs":1235.4209594726562,"jitter":0.00012305474736918189,"achievedTflops":354.4893079780885,"nominalPeakTflops":468.0,"mfuNominalPct":75.74557862779668,"configuredPeakTflops":468.0,"mfuConfiguredPct":75.74557862779668,"vramAllocatedGb":69.086945792,"vramAllocatedPct":67.74960220491786,"vramReservedGb":70.890029056,"vramReservedPct":69.51778246644118,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":101.032203264,"vramAllocatedPct":99.07648257646696,"vramReservedGb":101.160321024,"vramReservedPct":99.20212030984595,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 107.75 MiB is free. Including non-PyTorch memory, this process has 94.86 GiB memory in use. Of the allocated memory 94.09 GiB is allocated by PyTorch, and 122.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":26517.45515052299,"meanTps":26514.919423515752,"stepMs":1235.71435546875,"jitter":0.0003230152469057506,"achievedTflops":354.40514148511335,"nominalPeakTflops":468.0,"mfuNominalPct":75.72759433442593,"configuredPeakTflops":468.0,"mfuConfiguredPct":75.72759433442593,"vramAllocatedGb":69.086945792,"vramAllocatedPct":67.74960220491786,"vramReservedGb":70.890029056,"vramReservedPct":69.51778246644118,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"350m","modelLabel":"350M","parameters":348447744,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":101.032203264,"vramAllocatedPct":99.07648257646696,"vramReservedGb":101.160321024,"vramReservedPct":99.20212030984595,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 107.75 MiB is free. Including non-PyTorch memory, this process has 94.86 GiB memory in use. Of the allocated memory 94.09 GiB is allocated by PyTorch, and 122.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":29662.372747106645,"meanTps":29674.267937628396,"stepMs":552.349609375,"jitter":0.00288830011527459,"achievedTflops":96.26737507269554,"nominalPeakTflops":312.0,"mfuNominalPct":30.854927907915236,"configuredPeakTflops":312.0,"mfuConfiguredPct":30.854927907915236,"vramAllocatedGb":44.089475072,"vramAllocatedPct":51.88569524696823,"vramReservedGb":45.26702592,"vramReservedPct":53.271469160977446,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30951.128696450465,"meanTps":30955.365353067118,"stepMs":1058.7012939453125,"jitter":0.0008879824016955842,"achievedTflops":100.44995188171873,"nominalPeakTflops":312.0,"mfuNominalPct":32.19549739798678,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.19549739798678,"vramAllocatedGb":81.082891776,"vramAllocatedPct":95.42055571226835,"vramReservedGb":83.588284416,"vramReservedPct":98.36897001705995,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.162187776,"vramAllocatedPct":99.04435512403941,"vramReservedGb":84.343259136,"vramReservedPct":99.2574448328094,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 76.75 MiB is free. Process 764857 has 79.05 GiB memory in use. Of the allocated memory 78.38 GiB is allocated by PyTorch, and 172.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":29701.138136916885,"meanTps":29694.6832884633,"stepMs":551.6286926269531,"jitter":0.001750172775498514,"achievedTflops":96.39318572016191,"nominalPeakTflops":312.0,"mfuNominalPct":30.895251833385228,"configuredPeakTflops":312.0,"mfuConfiguredPct":30.895251833385228,"vramAllocatedGb":44.089475072,"vramAllocatedPct":51.88569524696823,"vramReservedGb":45.26702592,"vramReservedPct":53.271469160977446,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30949.32837932777,"meanTps":30955.972749818364,"stepMs":1058.7628784179688,"jitter":0.0010504915641072335,"achievedTflops":100.44410906512488,"nominalPeakTflops":312.0,"mfuNominalPct":32.19362470036054,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.19362470036054,"vramAllocatedGb":81.082891776,"vramAllocatedPct":95.42055571226835,"vramReservedGb":83.588284416,"vramReservedPct":98.36897001705995,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.162187776,"vramAllocatedPct":99.04435512403941,"vramReservedGb":84.343259136,"vramReservedPct":99.2574448328094,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 76.75 MiB is free. Process 852916 has 79.05 GiB memory in use. Of the allocated memory 78.38 GiB is allocated by PyTorch, and 172.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":101307.37026991224,"meanTps":101356.13263171355,"stepMs":323.4512939453125,"jitter":0.000512461642072179,"achievedTflops":328.7867324219161,"nominalPeakTflops":2250.0,"mfuNominalPct":14.612743663196271,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.612743663196271,"vramAllocatedGb":81.082891776,"vramAllocatedPct":42.340270647023516,"vramReservedGb":83.451969536,"vramReservedPct":43.577367540648794,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":108209.51113165637,"meanTps":108197.70488625816,"stepMs":605.6399230957031,"jitter":0.00035933923720859356,"achievedTflops":351.1871987907744,"nominalPeakTflops":2250.0,"mfuNominalPct":15.60831994625664,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.60831994625664,"vramAllocatedGb":155.069725696,"vramAllocatedPct":80.97508625206358,"vramReservedGb":160.195149824,"vramReservedPct":83.65150590122734,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.216674816,"vramAllocatedPct":99.3282962272219,"vramReservedGb":190.478024704,"vramReservedPct":99.4647692210818,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 245.81 MiB is free. Including non-PyTorch memory, this process has 178.10 GiB memory in use. Of the allocated memory 177.15 GiB is allocated by PyTorch, and 129.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":101035.09409555823,"meanTps":101049.00050860856,"stepMs":324.3229522705078,"jitter":0.0016147949249556417,"achievedTflops":327.90307713164765,"nominalPeakTflops":2250.0,"mfuNominalPct":14.573470094739896,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.573470094739896,"vramAllocatedGb":81.082891776,"vramAllocatedPct":42.340270647023516,"vramReservedGb":83.451969536,"vramReservedPct":43.577367540648794,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":108169.36787405254,"meanTps":108152.51331810767,"stepMs":605.8646850585938,"jitter":0.0004096142472155557,"achievedTflops":351.05691635958334,"nominalPeakTflops":2250.0,"mfuNominalPct":15.602529615981481,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.602529615981481,"vramAllocatedGb":155.069725696,"vramAllocatedPct":80.97508625206358,"vramReservedGb":160.195149824,"vramReservedPct":83.65150590122734,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.216674816,"vramAllocatedPct":99.3282962272219,"vramReservedGb":190.478024704,"vramReservedPct":99.4647692210818,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 245.81 MiB is free. Including non-PyTorch memory, this process has 178.10 GiB memory in use. Of the allocated memory 177.15 GiB is allocated by PyTorch, and 129.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":99126.05685452324,"meanTps":99135.63857970374,"stepMs":330.56898498535156,"jitter":0.0003044014658489786,"achievedTflops":321.70741619523835,"nominalPeakTflops":2250.0,"mfuNominalPct":14.298107386455039,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.298107386455039,"vramAllocatedGb":81.082891776,"vramAllocatedPct":28.20974679466498,"vramReservedGb":83.451969536,"vramReservedPct":29.033978420876586,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":108126.54310591241,"meanTps":107955.12400770074,"stepMs":606.1046447753906,"jitter":0.0006026447827954078,"achievedTflops":350.917931253702,"nominalPeakTflops":2250.0,"mfuNominalPct":15.596352500164535,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.596352500164535,"vramAllocatedGb":155.069725696,"vramAllocatedPct":53.9506867797373,"vramReservedGb":160.195149824,"vramReservedPct":55.733885598861605,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":277.92672,"vramAllocatedPct":96.69416355217379,"vramReservedGb":279.58181888,"vramReservedPct":97.26999304347905,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.60 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.84 GiB is free. Including non-PyTorch memory, this process has 259.83 GiB memory in use. Of the allocated memory 258.84 GiB is allocated by PyTorch, and 178.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":99392.70010020342,"meanTps":99396.5759412401,"stepMs":329.6821594238281,"jitter":0.0003850935825845075,"achievedTflops":322.5727901678919,"nominalPeakTflops":2250.0,"mfuNominalPct":14.336568451906308,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.336568451906308,"vramAllocatedGb":81.082891776,"vramAllocatedPct":28.20974679466498,"vramReservedGb":83.451969536,"vramReservedPct":29.033978420876586,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":108244.79528850545,"meanTps":108246.31975084294,"stepMs":605.4425048828125,"jitter":0.00038224938177763753,"achievedTflops":351.3017113144512,"nominalPeakTflops":2250.0,"mfuNominalPct":15.613409391753388,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.613409391753388,"vramAllocatedGb":155.069725696,"vramAllocatedPct":53.9506867797373,"vramReservedGb":160.195149824,"vramReservedPct":55.733885598861605,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":277.92672,"vramAllocatedPct":96.69416355217379,"vramReservedGb":279.58181888,"vramReservedPct":97.26999304347905,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.60 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.84 GiB is free. Including non-PyTorch memory, this process has 259.83 GiB memory in use. Of the allocated memory 258.84 GiB is allocated by PyTorch, and 178.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":19670.530333260434,"meanTps":19662.982106159976,"stepMs":104.11513900756836,"jitter":0.0010713247667666465,"achievedTflops":63.83947560821908,"nominalPeakTflops":165.2,"mfuNominalPct":38.64375036817136,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":31.06291895623344,"vramAllocatedGb":11.720235008,"vramAllocatedPct":46.39647213797791,"vramReservedGb":11.913920512,"vramReservedPct":47.163207965692315,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22760.162004834197,"meanTps":22760.887437298603,"stepMs":179.96356964111328,"jitter":0.000651790801199838,"achievedTflops":73.86668191095437,"nominalPeakTflops":165.2,"mfuNominalPct":44.713487839560756,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":35.94194237821152,"vramAllocatedGb":16.34441216,"vramAllocatedPct":64.70203565674673,"vramReservedGb":16.6723584,"vramReservedPct":66.0002646237025,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.022972928,"vramAllocatedPct":95.09887757067658,"vramReservedGb":24.152899584,"vramReservedPct":95.61321354354487,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 597.69 MiB is free. Including non-PyTorch memory, this process has 22.93 GiB memory in use. Of the allocated memory 22.37 GiB is allocated by PyTorch, and 103.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":19613.08336969202,"meanTps":19595.350262358144,"stepMs":104.42009353637695,"jitter":0.0018614623458262914,"achievedTflops":63.653035081839846,"nominalPeakTflops":165.2,"mfuNominalPct":38.530892906682716,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":30.972200996759312,"vramAllocatedGb":11.720235008,"vramAllocatedPct":46.39647213797791,"vramReservedGb":11.913920512,"vramReservedPct":47.163207965692315,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22712.160459340997,"meanTps":22710.04490648149,"stepMs":180.3439178466797,"jitter":0.001149486695977052,"achievedTflops":73.71089589803287,"nominalPeakTflops":165.2,"mfuNominalPct":44.61918637895453,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":35.8661402471979,"vramAllocatedGb":16.34441216,"vramAllocatedPct":64.70203565674673,"vramReservedGb":16.6723584,"vramReservedPct":66.0002646237025,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.022972928,"vramAllocatedPct":95.09887757067658,"vramReservedGb":24.152899584,"vramReservedPct":95.61321354354487,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 597.69 MiB is free. Including non-PyTorch memory, this process has 22.93 GiB memory in use. Of the allocated memory 22.37 GiB is allocated by PyTorch, and 103.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":33182.46486448534,"meanTps":33172.196246396925,"stepMs":123.43869018554688,"jitter":0.00029479073209094355,"achievedTflops":107.69161382268524,"nominalPeakTflops":209.5,"mfuNominalPct":51.404111609873624,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":40.041973024260784,"vramAllocatedGb":16.34441216,"vramAllocatedPct":48.54459937400973,"vramReservedGb":16.712204288,"vramReservedPct":49.63698014193785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":35300.64109082458,"meanTps":35300.481399330354,"stepMs":232.06377410888672,"jitter":0.00039700667352400197,"achievedTflops":114.56602225216454,"nominalPeakTflops":209.5,"mfuNominalPct":54.685452149004554,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":42.59802049276827,"vramAllocatedGb":25.592766464,"vramAllocatedPct":76.01317090546691,"vramReservedGb":26.273120256,"vramReservedPct":78.03389223468406,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.708504064,"vramAllocatedPct":97.14764962890222,"vramReservedGb":32.986103808,"vramReservedPct":97.97214966150588,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 108.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 101.88 MiB is free. Including non-PyTorch memory, this process has 31.25 GiB memory in use. Of the allocated memory 30.46 GiB is allocated by PyTorch, and 204.74 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":33163.967734881444,"meanTps":33151.10734032082,"stepMs":123.50753784179688,"jitter":0.0008195413067891908,"achievedTflops":107.63158254573622,"nominalPeakTflops":209.5,"mfuNominalPct":51.375457062403925,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":40.01965215184668,"vramAllocatedGb":16.34441216,"vramAllocatedPct":48.54459937400973,"vramReservedGb":16.712204288,"vramReservedPct":49.63698014193785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":35297.9453299945,"meanTps":35296.88663409035,"stepMs":232.0814971923828,"jitter":0.00019822011529095296,"achievedTflops":114.5572733290373,"nominalPeakTflops":209.5,"mfuNominalPct":54.68127605204645,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":42.594767461901554,"vramAllocatedGb":25.592766464,"vramAllocatedPct":76.01317090546691,"vramReservedGb":26.273120256,"vramReservedPct":78.03389223468406,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.708504064,"vramAllocatedPct":97.14764962890222,"vramReservedGb":32.986103808,"vramReservedPct":97.97214966150588,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 108.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 101.88 MiB is free. Including non-PyTorch memory, this process has 31.25 GiB memory in use. Of the allocated memory 30.46 GiB is allocated by PyTorch, and 204.74 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":64270.79619029149,"meanTps":64270.17355176955,"stepMs":254.92137908935547,"jitter":0.0012164106634367912,"achievedTflops":208.58684825458138,"nominalPeakTflops":989.5,"mfuNominalPct":21.0800250888915,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.0800250888915,"vramAllocatedGb":44.139544576,"vramAllocatedPct":51.91819090023311,"vramReservedGb":45.32994048,"vramReservedPct":53.31836850479162,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":68549.87228718399,"meanTps":68543.9563913741,"stepMs":478.0169372558594,"jitter":0.00016516811301880356,"achievedTflops":222.47432202804535,"nominalPeakTflops":989.5,"mfuNominalPct":22.483509047806503,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.483509047806503,"vramAllocatedGb":81.13296128,"vramAllocatedPct":95.43090243678233,"vramReservedGb":83.588284416,"vramReservedPct":98.31892351903699,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.876450816,"vramAllocatedPct":98.6578730553303,"vramReservedGb":84.070629376,"vramReservedPct":98.88627141429963,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 152.19 MiB is free. Including non-PyTorch memory, this process has 79.02 GiB memory in use. Of the allocated memory 78.12 GiB is allocated by PyTorch, and 185.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":64327.243431569295,"meanTps":64327.24747806739,"stepMs":254.69768524169922,"jitter":0.0003598166951660741,"achievedTflops":208.77004424480907,"nominalPeakTflops":989.5,"mfuNominalPct":21.098539084872062,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.098539084872062,"vramAllocatedGb":44.139544576,"vramAllocatedPct":51.91819090023311,"vramReservedGb":45.32994048,"vramReservedPct":53.31836850479162,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":68529.16958735585,"meanTps":68517.69595839351,"stepMs":478.1613464355469,"jitter":0.00043791814998744925,"achievedTflops":222.407132710914,"nominalPeakTflops":989.5,"mfuNominalPct":22.476718818687623,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.476718818687623,"vramAllocatedGb":81.13296128,"vramAllocatedPct":95.43090243678233,"vramReservedGb":83.588284416,"vramReservedPct":98.31892351903699,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.876450816,"vramAllocatedPct":98.6578730553303,"vramReservedGb":84.070629376,"vramReservedPct":98.88627141429963,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 152.19 MiB is free. Including non-PyTorch memory, this process has 79.02 GiB memory in use. Of the allocated memory 78.12 GiB is allocated by PyTorch, and 185.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":71402.06488031692,"meanTps":71384.87750098864,"stepMs":229.46115112304688,"jitter":0.0012381975993295902,"achievedTflops":231.73093465588966,"nominalPeakTflops":989.5,"mfuNominalPct":23.418992890943876,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.418992890943876,"vramAllocatedGb":44.139544576,"vramAllocatedPct":29.402692061374424,"vramReservedGb":45.30896896,"vramReservedPct":30.18168118285508,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":76109.83148747424,"meanTps":76002.258276985,"stepMs":430.53570556640625,"jitter":0.002781807226558596,"achievedTflops":247.00969666153975,"nominalPeakTflops":989.5,"mfuNominalPct":24.963082027442116,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.963082027442116,"vramAllocatedGb":81.13296128,"vramAllocatedPct":54.04513117338183,"vramReservedGb":83.567312896,"vramReservedPct":55.66672676576761,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.561457664,"vramAllocatedPct":94.96451945257286,"vramReservedGb":143.31936768,"vramReservedPct":95.46938634743421,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.80 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.17 GiB is free. Including non-PyTorch memory, this process has 133.63 GiB memory in use. Of the allocated memory 132.77 GiB is allocated by PyTorch, and 142.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":71373.23407232254,"meanTps":71374.75630902228,"stepMs":229.55384063720703,"jitter":0.0008120573724858025,"achievedTflops":231.63736607219923,"nominalPeakTflops":989.5,"mfuNominalPct":23.40953674302165,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.40953674302165,"vramAllocatedGb":44.139544576,"vramAllocatedPct":29.402692061374424,"vramReservedGb":45.30896896,"vramReservedPct":30.18168118285508,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":75861.51421936441,"meanTps":75826.71390210814,"stepMs":431.9449768066406,"jitter":0.0058464558113824295,"achievedTflops":246.20379850261764,"nominalPeakTflops":989.5,"mfuNominalPct":24.881637039173082,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.881637039173082,"vramAllocatedGb":81.13296128,"vramAllocatedPct":54.04513117338183,"vramReservedGb":83.567312896,"vramReservedPct":55.66672676576761,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.561457664,"vramAllocatedPct":94.96451945257286,"vramReservedGb":143.31936768,"vramReservedPct":95.46938634743421,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.80 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.17 GiB is free. Including non-PyTorch memory, this process has 133.63 GiB memory in use. Of the allocated memory 132.77 GiB is allocated by PyTorch, and 142.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":20811.930700927904,"meanTps":20797.81526579516,"stepMs":196.81018829345703,"jitter":0.002189102330740778,"achievedTflops":67.54381909547686,"nominalPeakTflops":364.2,"mfuNominalPct":18.545804254661412,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.962073963905585,"vramAllocatedGb":16.34441216,"vramAllocatedPct":32.125445532055764,"vramReservedGb":16.712204288,"vramReservedPct":32.8483522881702,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":20807.21049794403,"meanTps":20812.182419184253,"stepMs":393.7096710205078,"jitter":0.004713009737906795,"achievedTflops":67.5284999719886,"nominalPeakTflops":364.2,"mfuNominalPct":18.541598015373037,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.958680524479695,"vramAllocatedGb":25.592766464,"vramAllocatedPct":50.303370779280165,"vramReservedGb":26.294091776,"vramReservedPct":51.68184728185192,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":20190.52828263728,"meanTps":20194.253439989145,"stepMs":811.4696044921875,"jitter":0.0022796933280321394,"achievedTflops":65.52709642184993,"nominalPeakTflops":364.2,"mfuNominalPct":17.992063817092237,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.515336509441564,"vramAllocatedGb":44.089475072,"vramAllocatedPct":86.65922127372897,"vramReservedGb":45.422215168,"vramReservedPct":89.27876298274292,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14560.007185748787,"meanTps":14557.044058330504,"stepMs":562.6370849609375,"jitter":0.0009854250146936847,"achievedTflops":47.253592447298146,"nominalPeakTflops":154.8,"mfuNominalPct":30.52557651634247,"configuredPeakTflops":180.6,"mfuConfiguredPct":26.16477987115069,"vramAllocatedGb":25.592766464,"vramAllocatedPct":50.1485297592447,"vramReservedGb":26.247954432,"vramReservedPct":51.43235788143559,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":15504.487505749881,"meanTps":15502.264160638932,"stepMs":1056.726318359375,"jitter":0.0006323146158895154,"achievedTflops":50.31884423917288,"nominalPeakTflops":154.8,"mfuNominalPct":32.50571333279902,"configuredPeakTflops":180.6,"mfuConfiguredPct":27.86203999954202,"vramAllocatedGb":44.089475072,"vramAllocatedPct":86.39247170984031,"vramReservedGb":45.273317376,"vramReservedPct":88.71218775522783,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.54943488,"vramAllocatedPct":99.05063772447208,"vramReservedGb":50.660900864,"vramReservedPct":99.26905315930327,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 38.25 MiB is free. Process 1415479 has 47.48 GiB memory in use. Of the allocated memory 47.08 GiB is allocated by PyTorch, and 86.30 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14589.318779130963,"meanTps":14583.075372214662,"stepMs":561.5066833496094,"jitter":0.0012856298266725457,"achievedTflops":47.34872138988672,"nominalPeakTflops":154.8,"mfuNominalPct":30.587029321632247,"configuredPeakTflops":180.6,"mfuConfiguredPct":26.217453704256215,"vramAllocatedGb":25.592766464,"vramAllocatedPct":50.1485297592447,"vramReservedGb":26.247954432,"vramReservedPct":51.43235788143559,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":15521.50401317234,"meanTps":15514.74368324123,"stepMs":1055.5678100585938,"jitter":0.0007295372618097753,"achievedTflops":50.37407024945976,"nominalPeakTflops":154.8,"mfuNominalPct":32.5413890500386,"configuredPeakTflops":180.6,"mfuConfiguredPct":27.892619185747378,"vramAllocatedGb":44.089475072,"vramAllocatedPct":86.39247170984031,"vramReservedGb":45.273317376,"vramReservedPct":88.71218775522783,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.54943488,"vramAllocatedPct":99.05063772447208,"vramReservedGb":50.660900864,"vramReservedPct":99.26905315930327,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 38.25 MiB is free. Process 1449495 has 47.48 GiB memory in use. Of the allocated memory 47.08 GiB is allocated by PyTorch, and 86.30 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":36298.68642547252,"meanTps":36299.5212446311,"stepMs":112.84154891967773,"jitter":0.00039164126246355943,"achievedTflops":117.8051159480483,"nominalPeakTflops":468.0,"mfuNominalPct":25.172033322232544,"configuredPeakTflops":468.0,"mfuConfiguredPct":25.172033322232544,"vramAllocatedGb":16.34441216,"vramAllocatedPct":16.02802684963165,"vramReservedGb":16.712204288,"vramReservedPct":16.388699478598635,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":43538.72579679516,"meanTps":43535.67675221597,"stepMs":188.15433502197266,"jitter":0.0002687846794420432,"achievedTflops":141.30221079081284,"nominalPeakTflops":468.0,"mfuNominalPct":30.192780083507017,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.192780083507017,"vramAllocatedGb":25.592766464,"vramAllocatedPct":25.0973570677102,"vramReservedGb":26.273120256,"vramReservedPct":25.764540979782122,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":41600.50211558809,"meanTps":41598.60420714305,"stepMs":393.8414001464844,"jitter":0.0002507303728401959,"achievedTflops":135.0118270887288,"nominalPeakTflops":468.0,"mfuNominalPct":28.848681001865128,"configuredPeakTflops":468.0,"mfuConfiguredPct":28.848681001865128,"vramAllocatedGb":44.089475072,"vramAllocatedPct":43.23601750386729,"vramReservedGb":45.443186688,"vramReservedPct":44.563524783756286,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":36240.65993031344,"meanTps":36240.479755984736,"stepMs":113.02222442626953,"jitter":0.00043990460001853645,"achievedTflops":117.61679458814704,"nominalPeakTflops":468.0,"mfuNominalPct":25.131793715416034,"configuredPeakTflops":468.0,"mfuConfiguredPct":25.131793715416034,"vramAllocatedGb":16.470900736,"vramAllocatedPct":16.152066935775665,"vramReservedGb":17.884512256,"vramReservedPct":17.538314613312732,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":43359.63658956891,"meanTps":43365.22718796503,"stepMs":188.9314727783203,"jitter":0.00047387159699472113,"achievedTflops":140.72098797258087,"nominalPeakTflops":468.0,"mfuNominalPct":30.068587173628394,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.068587173628394,"vramAllocatedGb":25.656836096,"vramAllocatedPct":25.160186478268944,"vramReservedGb":27.762098176,"vramReservedPct":27.224696159830437,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":41475.45447497778,"meanTps":41473.46639706294,"stepMs":395.02882385253906,"jitter":0.00017449690711463097,"achievedTflops":134.60599279410843,"nominalPeakTflops":468.0,"mfuNominalPct":28.76196427224539,"configuredPeakTflops":468.0,"mfuConfiguredPct":28.76196427224539,"vramAllocatedGb":44.161695744,"vramAllocatedPct":43.30684017149111,"vramReservedGb":45.789216768,"vramReservedPct":44.90285662137315,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":29037.482774117412,"meanTps":29040.179478790164,"stepMs":564.2362365722656,"jitter":0.00135125951564542,"achievedTflops":100.40505528478968,"nominalPeakTflops":312.0,"mfuNominalPct":32.181107463073616,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.181107463073616,"vramAllocatedGb":44.096364544,"vramAllocatedPct":51.8938029652847,"vramReservedGb":45.290094592,"vramReservedPct":53.29861700256979,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30189.378577380736,"meanTps":30179.242483918715,"stepMs":1085.4148559570312,"jitter":0.0009968516322238496,"achievedTflops":104.38805073617439,"nominalPeakTflops":312.0,"mfuNominalPct":33.457708569286666,"configuredPeakTflops":312.0,"mfuConfiguredPct":33.457708569286666,"vramAllocatedGb":81.09490176,"vramAllocatedPct":95.43468938858742,"vramReservedGb":83.609255936,"vramReservedPct":98.393649873053,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.163957248,"vramAllocatedPct":99.04643748688882,"vramReservedGb":84.341161984,"vramReservedPct":99.2549768472101,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 78.75 MiB is free. Process 764857 has 79.05 GiB memory in use. Of the allocated memory 78.38 GiB is allocated by PyTorch, and 169.00 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":29006.833156936227,"meanTps":29016.33222671654,"stepMs":564.8324279785156,"jitter":0.0008760178186922293,"achievedTflops":100.29907583293866,"nominalPeakTflops":312.0,"mfuNominalPct":32.14713969004444,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.14713969004444,"vramAllocatedGb":44.096364544,"vramAllocatedPct":51.8938029652847,"vramReservedGb":45.290094592,"vramReservedPct":53.29861700256979,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30203.441353570044,"meanTps":30211.04478176681,"stepMs":1084.9094848632812,"jitter":0.0008123992378801492,"achievedTflops":104.4366766391746,"nominalPeakTflops":312.0,"mfuNominalPct":33.473293794607244,"configuredPeakTflops":312.0,"mfuConfiguredPct":33.473293794607244,"vramAllocatedGb":81.09490176,"vramAllocatedPct":95.43468938858742,"vramReservedGb":83.609255936,"vramReservedPct":98.393649873053,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.163957248,"vramAllocatedPct":99.04643748688882,"vramReservedGb":84.341161984,"vramReservedPct":99.2549768472101,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 78.75 MiB is free. Process 852916 has 79.05 GiB memory in use. Of the allocated memory 78.38 GiB is allocated by PyTorch, and 169.00 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":97445.64535804208,"meanTps":97445.93763714713,"stepMs":336.26951599121094,"jitter":0.0012990888048099627,"achievedTflops":336.94502672791026,"nominalPeakTflops":2250.0,"mfuNominalPct":14.975334521240455,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.975334521240455,"vramAllocatedGb":81.09490176,"vramAllocatedPct":42.34654208063779,"vramReservedGb":83.475038208,"vramReservedPct":43.58941365536614,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":103884.53969075116,"meanTps":103833.9247592716,"stepMs":630.8542175292969,"jitter":0.0003284094863198137,"achievedTflops":359.20926865541054,"nominalPeakTflops":2250.0,"mfuNominalPct":15.964856384684914,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.964856384684914,"vramAllocatedGb":155.09197568,"vramAllocatedPct":80.98670486017951,"vramReservedGb":160.218218496,"vramReservedPct":83.66355201594467,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.218444288,"vramAllocatedPct":99.32922021897578,"vramReservedGb":190.486413312,"vramReservedPct":99.46914962643356,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 237.81 MiB is free. Including non-PyTorch memory, this process has 178.11 GiB memory in use. Of the allocated memory 177.15 GiB is allocated by PyTorch, and 135.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":97162.43479378219,"meanTps":97102.74308543284,"stepMs":337.2496795654297,"jitter":0.001962959624233016,"achievedTflops":335.9657485796302,"nominalPeakTflops":2250.0,"mfuNominalPct":14.931811047983565,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.931811047983565,"vramAllocatedGb":81.09490176,"vramAllocatedPct":42.34654208063779,"vramReservedGb":83.475038208,"vramReservedPct":43.58941365536614,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":103894.21954285711,"meanTps":103886.67302343456,"stepMs":630.7954406738281,"jitter":0.00027478793927693425,"achievedTflops":359.24273939712083,"nominalPeakTflops":2250.0,"mfuNominalPct":15.96634397320537,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.96634397320537,"vramAllocatedGb":155.09197568,"vramAllocatedPct":80.98670486017951,"vramReservedGb":160.218218496,"vramReservedPct":83.66355201594467,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.218444288,"vramAllocatedPct":99.32922021897578,"vramReservedGb":190.486413312,"vramReservedPct":99.46914962643356,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 237.81 MiB is free. Including non-PyTorch memory, this process has 178.11 GiB memory in use. Of the allocated memory 177.15 GiB is allocated by PyTorch, and 135.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":95451.06954237854,"meanTps":95450.65461824142,"stepMs":343.2963104248047,"jitter":0.0003240224769312232,"achievedTflops":330.0482341719139,"nominalPeakTflops":2250.0,"mfuNominalPct":14.668810407640619,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.668810407640619,"vramAllocatedGb":81.09490176,"vramAllocatedPct":28.213925217513832,"vramReservedGb":83.475038208,"vramReservedPct":29.042004298860896,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":103890.32929203729,"meanTps":103887.20523951936,"stepMs":630.8190612792969,"jitter":0.00046032966793205,"achievedTflops":359.22928778867123,"nominalPeakTflops":2250.0,"mfuNominalPct":15.965746123940942,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.965746123940942,"vramAllocatedGb":155.09197568,"vramAllocatedPct":53.95842782598118,"vramReservedGb":160.218218496,"vramReservedPct":55.741911476845914,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":277.944874496,"vramAllocatedPct":96.70047972719082,"vramReservedGb":279.611179008,"vramReservedPct":97.28020779727727,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.61 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.81 GiB is free. Including non-PyTorch memory, this process has 259.86 GiB memory in use. Of the allocated memory 258.86 GiB is allocated by PyTorch, and 189.11 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":95690.47578201548,"meanTps":95534.9072439518,"stepMs":342.4374237060547,"jitter":0.00039618396398998493,"achievedTflops":330.8760468619209,"nominalPeakTflops":2250.0,"mfuNominalPct":14.70560208275204,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.70560208275204,"vramAllocatedGb":81.09490176,"vramAllocatedPct":28.213925217513832,"vramReservedGb":83.475038208,"vramReservedPct":29.042004298860896,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":104012.82087049248,"meanTps":103979.25655412994,"stepMs":630.076171875,"jitter":0.00024752683478560065,"achievedTflops":359.65283599367194,"nominalPeakTflops":2250.0,"mfuNominalPct":15.98457048860764,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.98457048860764,"vramAllocatedGb":155.09197568,"vramAllocatedPct":53.95842782598118,"vramReservedGb":160.218218496,"vramReservedPct":55.741911476845914,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":277.944874496,"vramAllocatedPct":96.70047972719082,"vramReservedGb":279.611179008,"vramReservedPct":97.28020779727727,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.61 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.81 GiB is free. Including non-PyTorch memory, this process has 259.86 GiB memory in use. Of the allocated memory 258.86 GiB is allocated by PyTorch, and 189.11 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":19277.83424473224,"meanTps":19271.680608341816,"stepMs":106.23600006103516,"jitter":0.0006586995644029095,"achievedTflops":66.65839556997152,"nominalPeakTflops":165.2,"mfuNominalPct":40.35011838376,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":32.43454492091713,"vramAllocatedGb":11.72264448,"vramAllocatedPct":46.406010427989926,"vramReservedGb":11.916017664,"vramReservedPct":47.17150988577077,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22281.13921598346,"meanTps":22282.267972515117,"stepMs":183.83261108398438,"jitter":0.0007939915273131183,"achievedTflops":77.04314565389913,"nominalPeakTflops":165.2,"mfuNominalPct":46.63628671543531,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":37.487541474608285,"vramAllocatedGb":16.347461632,"vramAllocatedPct":64.71410749170457,"vramReservedGb":16.674455552,"vramReservedPct":66.00856654378096,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.0257664,"vramAllocatedPct":95.10993598765609,"vramReservedGb":24.154996736,"vramReservedPct":95.62151546362333,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 595.69 MiB is free. Including non-PyTorch memory, this process has 22.94 GiB memory in use. Of the allocated memory 22.38 GiB is allocated by PyTorch, and 103.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":19224.096901000063,"meanTps":19223.075551660906,"stepMs":106.53296279907227,"jitter":0.0018216771689968448,"achievedTflops":66.47258397568635,"nominalPeakTflops":165.2,"mfuNominalPct":40.2376416317714,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":32.34413298630428,"vramAllocatedGb":11.72264448,"vramAllocatedPct":46.406010427989926,"vramReservedGb":11.916017664,"vramReservedPct":47.17150988577077,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22249.137617895427,"meanTps":22240.193139346833,"stepMs":184.0970230102539,"jitter":0.001326661650773219,"achievedTflops":76.93249135751175,"nominalPeakTflops":165.2,"mfuNominalPct":46.56930469583036,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":37.433699468418666,"vramAllocatedGb":16.347461632,"vramAllocatedPct":64.71410749170457,"vramReservedGb":16.674455552,"vramReservedPct":66.00856654378096,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.0257664,"vramAllocatedPct":95.10993598765609,"vramReservedGb":24.154996736,"vramReservedPct":95.62151546362333,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 595.69 MiB is free. Including non-PyTorch memory, this process has 22.94 GiB memory in use. Of the allocated memory 22.38 GiB is allocated by PyTorch, and 103.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":32383.739574543037,"meanTps":32384.33926969811,"stepMs":126.48323059082031,"jitter":0.0002593904542112301,"achievedTflops":111.97565531432514,"nominalPeakTflops":209.5,"mfuNominalPct":53.449000150035864,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":41.63486840166224,"vramAllocatedGb":16.347461632,"vramAllocatedPct":48.55365662214402,"vramReservedGb":16.71430144,"vramReservedPct":49.64320890089655,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":34538.26972356886,"meanTps":34538.439480209556,"stepMs":237.18617248535156,"jitter":0.00020721180002898824,"achievedTflops":119.42553381820517,"nominalPeakTflops":209.5,"mfuNominalPct":57.00502807551559,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":44.40488756562007,"vramAllocatedGb":25.597095936,"vramAllocatedPct":76.02602988636409,"vramReservedGb":26.275217408,"vramReservedPct":78.04012099364277,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.710273536,"vramAllocatedPct":97.15290514427363,"vramReservedGb":32.986103808,"vramReservedPct":97.97214966150588,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 108.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 101.88 MiB is free. Including non-PyTorch memory, this process has 31.25 GiB memory in use. Of the allocated memory 30.46 GiB is allocated by PyTorch, and 203.05 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":32381.171099038136,"meanTps":32366.46646489471,"stepMs":126.4932632446289,"jitter":0.00037561909500911505,"achievedTflops":111.96677410629917,"nominalPeakTflops":209.5,"mfuNominalPct":53.444760909928,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":41.631566184529675,"vramAllocatedGb":16.347461632,"vramAllocatedPct":48.55365662214402,"vramReservedGb":16.71430144,"vramReservedPct":49.64320890089655,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":34527.85088305309,"meanTps":34526.271335561294,"stepMs":237.25774383544922,"jitter":0.00036934075211058823,"achievedTflops":119.3895078215261,"nominalPeakTflops":209.5,"mfuNominalPct":56.98783189571651,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":44.391492353718334,"vramAllocatedGb":25.597095936,"vramAllocatedPct":76.02602988636409,"vramReservedGb":26.275217408,"vramReservedPct":78.04012099364277,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.710273536,"vramAllocatedPct":97.15290514427363,"vramReservedGb":32.986103808,"vramReservedPct":97.97214966150588,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 108.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 101.88 MiB is free. Including non-PyTorch memory, this process has 31.25 GiB memory in use. Of the allocated memory 30.46 GiB is allocated by PyTorch, and 203.05 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":62628.4409887734,"meanTps":62624.47852294434,"stepMs":261.60638427734375,"jitter":0.0003493631477507017,"achievedTflops":216.55499992179026,"nominalPeakTflops":989.5,"mfuNominalPct":21.88529559593636,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.88529559593636,"vramAllocatedGb":44.146434048,"vramAllocatedPct":51.926294493642004,"vramReservedGb":45.332037632,"vramReservedPct":53.320835234771025,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":66694.37927480064,"meanTps":66662.32061242462,"stepMs":491.3157653808594,"jitter":0.00047154780897590504,"achievedTflops":230.61409593809483,"nominalPeakTflops":989.5,"mfuNominalPct":23.306123894703873,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.306123894703873,"vramAllocatedGb":81.144971264,"vramAllocatedPct":95.44502892240901,"vramReservedGb":83.609255936,"vramReservedPct":98.34359081883102,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.878220288,"vramAllocatedPct":98.65995435875041,"vramReservedGb":84.068532224,"vramReservedPct":98.88380468432022,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 154.19 MiB is free. Including non-PyTorch memory, this process has 79.02 GiB memory in use. Of the allocated memory 78.12 GiB is allocated by PyTorch, and 181.50 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":62591.07173752157,"meanTps":62578.54097972171,"stepMs":261.7625732421875,"jitter":0.001095688898174233,"achievedTflops":216.42578549342264,"nominalPeakTflops":989.5,"mfuNominalPct":21.872237038243824,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.872237038243824,"vramAllocatedGb":44.146434048,"vramAllocatedPct":51.926294493642004,"vramReservedGb":45.332037632,"vramReservedPct":53.320835234771025,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":66712.62851248935,"meanTps":66704.34011529418,"stepMs":491.1813659667969,"jitter":0.00017967106121137634,"achievedTflops":230.67719767915463,"nominalPeakTflops":989.5,"mfuNominalPct":23.31250102871699,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.31250102871699,"vramAllocatedGb":81.144971264,"vramAllocatedPct":95.44502892240901,"vramReservedGb":83.609255936,"vramReservedPct":98.34359081883102,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.878220288,"vramAllocatedPct":98.65995435875041,"vramReservedGb":84.068532224,"vramReservedPct":98.88380468432022,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 154.19 MiB is free. Including non-PyTorch memory, this process has 79.02 GiB memory in use. Of the allocated memory 78.12 GiB is allocated by PyTorch, and 181.50 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":69503.29148734034,"meanTps":69457.64210607305,"stepMs":235.72984313964844,"jitter":0.001418273060979335,"achievedTflops":240.32667977961006,"nominalPeakTflops":989.5,"mfuNominalPct":24.28768870940981,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.28768870940981,"vramAllocatedGb":44.146434048,"vramAllocatedPct":29.407281348047572,"vramReservedGb":45.332037632,"vramReservedPct":30.19704792634091,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":73587.17789430512,"meanTps":73560.9546399935,"stepMs":445.2949676513672,"jitter":0.0029440299510147244,"achievedTflops":254.44783634327723,"nominalPeakTflops":989.5,"mfuNominalPct":25.71478891796637,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.71478891796637,"vramAllocatedGb":81.144971264,"vramAllocatedPct":54.05313138871269,"vramReservedGb":83.588284416,"vramReservedPct":55.68069653257291,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.571419648,"vramAllocatedPct":94.97115543286414,"vramReservedGb":143.31936768,"vramReservedPct":95.46938634743421,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.17 GiB is free. Including non-PyTorch memory, this process has 133.63 GiB memory in use. Of the allocated memory 132.78 GiB is allocated by PyTorch, and 133.30 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":69276.78798944585,"meanTps":69286.90254497586,"stepMs":236.50057220458984,"jitter":0.0007878700146584025,"achievedTflops":239.54348185556117,"nominalPeakTflops":989.5,"mfuNominalPct":24.208537832800523,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.208537832800523,"vramAllocatedGb":44.146434048,"vramAllocatedPct":29.407281348047572,"vramReservedGb":45.332037632,"vramReservedPct":30.19704792634091,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":73471.10086754037,"meanTps":73478.16638597281,"stepMs":445.9984893798828,"jitter":0.0037122625894588025,"achievedTflops":254.0464682088466,"nominalPeakTflops":989.5,"mfuNominalPct":25.67422619594205,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.67422619594205,"vramAllocatedGb":81.144971264,"vramAllocatedPct":54.05313138871269,"vramReservedGb":83.588284416,"vramReservedPct":55.68069653257291,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.571419648,"vramAllocatedPct":94.97115543286414,"vramReservedGb":143.31936768,"vramReservedPct":95.46938634743421,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.17 GiB is free. Including non-PyTorch memory, this process has 133.63 GiB memory in use. Of the allocated memory 132.78 GiB is allocated by PyTorch, and 133.30 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":20464.50938187925,"meanTps":20466.24057856245,"stepMs":200.1513900756836,"jitter":0.0022173237855576773,"achievedTflops":70.76165010057889,"nominalPeakTflops":364.2,"mfuNominalPct":19.429338303289097,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.674876795407146,"vramAllocatedGb":16.347461632,"vramAllocatedPct":32.131439363199924,"vramReservedGb":16.71430144,"vramReservedPct":32.852474305021516,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":20488.598139500908,"meanTps":20478.244681709297,"stepMs":399.83213806152344,"jitter":0.00370890481040265,"achievedTflops":70.84494358230246,"nominalPeakTflops":364.2,"mfuNominalPct":19.452208561862292,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.693327680343009,"vramAllocatedGb":25.597095936,"vramAllocatedPct":50.311880489850175,"vramReservedGb":26.296188928,"vramReservedPct":51.68596929870324,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19827.520109459452,"meanTps":19829.61452882329,"stepMs":826.3262329101562,"jitter":0.0016222216682477099,"achievedTflops":68.55908510516767,"nominalPeakTflops":364.2,"mfuNominalPct":18.824570319925225,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.186972190471074,"vramAllocatedGb":44.096364544,"vramAllocatedPct":86.67276274315068,"vramReservedGb":45.42431232,"vramReservedPct":89.28288499959424,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14307.652570373533,"meanTps":14306.421839588422,"stepMs":572.5607299804688,"jitter":0.0008365355322847059,"achievedTflops":49.472630203482986,"nominalPeakTflops":154.8,"mfuNominalPct":31.959063438942493,"configuredPeakTflops":180.6,"mfuConfiguredPct":27.393482947665,"vramAllocatedGb":25.597095936,"vramAllocatedPct":50.157013275699995,"vramReservedGb":26.250051584,"vramReservedPct":51.436467210125386,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":15214.020985774614,"meanTps":15216.984269524857,"stepMs":1076.9013671875,"jitter":0.0007812156936269153,"achievedTflops":52.60664741718751,"nominalPeakTflops":154.8,"mfuNominalPct":33.98362236252423,"configuredPeakTflops":180.6,"mfuConfiguredPct":29.128819167877914,"vramAllocatedGb":44.096364544,"vramAllocatedPct":86.40597149666887,"vramReservedGb":45.296386048,"vramReservedPct":88.75739037081554,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.551208448,"vramAllocatedPct":99.0541129965867,"vramReservedGb":50.669289472,"vramReservedPct":99.28549047406243,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 30.25 MiB is free. Process 1415479 has 47.49 GiB memory in use. Of the allocated memory 47.08 GiB is allocated by PyTorch, and 92.61 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14333.386960743508,"meanTps":14335.171433828236,"stepMs":571.5327453613281,"jitter":0.0015993694890293856,"achievedTflops":49.561613911468875,"nominalPeakTflops":154.8,"mfuNominalPct":32.016546454437254,"configuredPeakTflops":180.6,"mfuConfiguredPct":27.442754103803367,"vramAllocatedGb":25.597095936,"vramAllocatedPct":50.157013275699995,"vramReservedGb":26.250051584,"vramReservedPct":51.436467210125386,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":15245.547314985586,"meanTps":15247.172788409254,"stepMs":1074.6744384765625,"jitter":0.0005250051881400764,"achievedTflops":52.71565834117076,"nominalPeakTflops":154.8,"mfuNominalPct":34.05404285605346,"configuredPeakTflops":180.6,"mfuConfiguredPct":29.189179590902967,"vramAllocatedGb":44.096364544,"vramAllocatedPct":86.40597149666887,"vramReservedGb":45.296386048,"vramReservedPct":88.75739037081554,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.551208448,"vramAllocatedPct":99.0541129965867,"vramReservedGb":50.669289472,"vramReservedPct":99.28549047406243,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 30.25 MiB is free. Process 1449495 has 47.49 GiB memory in use. Of the allocated memory 47.08 GiB is allocated by PyTorch, and 92.61 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":35614.61741831578,"meanTps":35614.628286191335,"stepMs":115.00895690917969,"jitter":0.0002662875550915476,"achievedTflops":123.14730097816883,"nominalPeakTflops":468.0,"mfuNominalPct":26.313525850036072,"configuredPeakTflops":468.0,"mfuConfiguredPct":26.313525850036072,"vramAllocatedGb":16.347461632,"vramAllocatedPct":16.03101729178489,"vramReservedGb":16.71430144,"vramReservedPct":16.390756035190254,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":42833.62467733419,"meanTps":42833.72224521709,"stepMs":191.25161743164062,"jitter":0.00034315211284371473,"achievedTflops":148.10899716173452,"nominalPeakTflops":468.0,"mfuNominalPct":31.64722161575524,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.64722161575524,"vramAllocatedGb":25.597095936,"vramAllocatedPct":25.1016027323925,"vramReservedGb":26.275217408,"vramReservedPct":25.766597536373737,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":41098.03077405702,"meanTps":41095.86951886484,"stepMs":398.6565704345703,"jitter":0.0002949156746193126,"achievedTflops":142.10770554957668,"nominalPeakTflops":468.0,"mfuNominalPct":30.364894348200146,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.364894348200146,"vramAllocatedGb":44.096364544,"vramAllocatedPct":43.242773613607724,"vramReservedGb":45.44528384,"vramReservedPct":44.565581340347904,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":35577.21809449719,"meanTps":35583.594717470005,"stepMs":115.12985610961914,"jitter":0.0003999698084209788,"achievedTflops":123.01798256565938,"nominalPeakTflops":468.0,"mfuNominalPct":26.28589371061098,"configuredPeakTflops":468.0,"mfuConfiguredPct":26.28589371061098,"vramAllocatedGb":16.472670208,"vramAllocatedPct":16.15380215539984,"vramReservedGb":17.886609408,"vramReservedPct":17.54037116990435,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":42648.3571985891,"meanTps":42651.02332927649,"stepMs":192.08242797851562,"jitter":0.0002890809444378334,"achievedTflops":147.46838407586282,"nominalPeakTflops":468.0,"mfuNominalPct":31.510338477748466,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.510338477748466,"vramAllocatedGb":25.662701568,"vramAllocatedPct":25.165938409986126,"vramReservedGb":27.764195328,"vramReservedPct":27.226752716422055,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":40943.73146412162,"meanTps":40916.69536319104,"stepMs":400.158935546875,"jitter":0.0009068091840949427,"achievedTflops":141.5741733951202,"nominalPeakTflops":468.0,"mfuNominalPct":30.25089175109406,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.25089175109406,"vramAllocatedGb":44.169609216,"vramAllocatedPct":43.31460045925479,"vramReservedGb":45.797605376,"vramReservedPct":44.91108284773962,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":19021.837395046932,"meanTps":18812.443886805002,"stepMs":107.66572952270508,"jitter":0.004870519936543199,"achievedTflops":73.85128237674925,"nominalPeakTflops":312.0,"mfuNominalPct":23.670282813060656,"configuredPeakTflops":312.0,"mfuConfiguredPct":23.670282813060656,"vramAllocatedGb":11.72648704,"vramAllocatedPct":13.80004937417284,"vramReservedGb":11.792285696,"vramReservedPct":13.87748302488655,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22484.85339638312,"meanTps":22456.834226886152,"stepMs":182.16707611083984,"jitter":0.007637341872428461,"achievedTflops":87.29625970877468,"nominalPeakTflops":312.0,"mfuNominalPct":27.979570419479067,"configuredPeakTflops":312.0,"mfuConfiguredPct":27.979570419479067,"vramAllocatedGb":16.351640576,"vramAllocatedPct":19.243056042939866,"vramReservedGb":16.697524224,"vramReservedPct":19.650101341658672,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25847.22178825573,"meanTps":25839.919384324527,"stepMs":316.9392852783203,"jitter":0.0026344411587096257,"achievedTflops":100.35047799514794,"nominalPeakTflops":312.0,"mfuNominalPct":32.163614742034596,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.163614742034596,"vramAllocatedGb":25.60191488,"vramAllocatedPct":30.129030818198927,"vramReservedGb":26.254245888,"vramReservedPct":30.896711717687126,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":27653.253605509526,"meanTps":27652.96759511604,"stepMs":592.4800109863281,"jitter":0.0016175857849733586,"achievedTflops":107.36230145612093,"nominalPeakTflops":312.0,"mfuNominalPct":34.41099405644901,"configuredPeakTflops":312.0,"mfuConfiguredPct":34.41099405644901,"vramAllocatedGb":44.102463488,"vramAllocatedPct":51.90098036871705,"vramReservedGb":45.296386048,"vramReservedPct":53.3060209593677,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":28809.164654772223,"meanTps":28805.595685770724,"stepMs":1137.4158325195312,"jitter":0.000740845683067745,"achievedTflops":111.85006525772573,"nominalPeakTflops":312.0,"mfuNominalPct":35.84937989029671,"configuredPeakTflops":312.0,"mfuConfiguredPct":35.84937989029671,"vramAllocatedGb":81.103560704,"vramAllocatedPct":95.44487946975329,"vramReservedGb":83.554729984,"vramReservedPct":98.32948224747109,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.167496192,"vramAllocatedPct":99.05060221258765,"vramReservedGb":84.393590784,"vramReservedPct":99.3166764871927,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 68.75 MiB is free. Process 703773 has 79.06 GiB memory in use. Of the allocated memory 78.39 GiB is allocated by PyTorch, and 175.62 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":18884.242467112854,"meanTps":18820.434538054924,"stepMs":108.4502067565918,"jitter":0.01017830949330892,"achievedTflops":73.31707731204219,"nominalPeakTflops":312.0,"mfuNominalPct":23.49906324103916,"configuredPeakTflops":312.0,"mfuConfiguredPct":23.49906324103916,"vramAllocatedGb":11.72648704,"vramAllocatedPct":13.80004937417284,"vramReservedGb":11.792285696,"vramReservedPct":13.87748302488655,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22341.31581708125,"meanTps":22331.194186563665,"stepMs":183.33745574951172,"jitter":0.006961315328533553,"achievedTflops":86.73898261295335,"nominalPeakTflops":312.0,"mfuNominalPct":27.800955965690175,"configuredPeakTflops":312.0,"mfuConfiguredPct":27.800955965690175,"vramAllocatedGb":16.351640576,"vramAllocatedPct":19.243056042939866,"vramReservedGb":16.697524224,"vramReservedPct":19.650101341658672,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25817.820377966633,"meanTps":25821.219359814822,"stepMs":317.3002166748047,"jitter":0.0032456002708781857,"achievedTflops":100.23632856739076,"nominalPeakTflops":312.0,"mfuNominalPct":32.127028386984215,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.127028386984215,"vramAllocatedGb":25.60191488,"vramAllocatedPct":30.129030818198927,"vramReservedGb":26.254245888,"vramReservedPct":30.896711717687126,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":27605.196434031906,"meanTps":27596.043198695643,"stepMs":593.5114440917969,"jitter":0.00143997161071551,"achievedTflops":107.1757220175886,"nominalPeakTflops":312.0,"mfuNominalPct":34.35119295435532,"configuredPeakTflops":312.0,"mfuConfiguredPct":34.35119295435532,"vramAllocatedGb":44.102463488,"vramAllocatedPct":51.90098036871705,"vramReservedGb":45.296386048,"vramReservedPct":53.3060209593677,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":28715.700128732893,"meanTps":28721.773216683552,"stepMs":1141.117919921875,"jitter":0.0008758691930862925,"achievedTflops":111.48719415535069,"nominalPeakTflops":312.0,"mfuNominalPct":35.73307504979189,"configuredPeakTflops":312.0,"mfuConfiguredPct":35.73307504979189,"vramAllocatedGb":81.103560704,"vramAllocatedPct":95.44487946975329,"vramReservedGb":83.554729984,"vramReservedPct":98.32948224747109,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.167496192,"vramAllocatedPct":99.05060221258765,"vramReservedGb":84.393590784,"vramReservedPct":99.3166764871927,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 68.75 MiB is free. Process 736416 has 79.06 GiB memory in use. Of the allocated memory 78.39 GiB is allocated by PyTorch, and 175.62 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":12234.105420360112,"meanTps":12216.079361261713,"stepMs":167.40087890625,"jitter":0.015028394898669046,"achievedTflops":47.49827029123881,"nominalPeakTflops":2250.0,"mfuNominalPct":2.1110342351661697,"configuredPeakTflops":2250.0,"mfuConfiguredPct":2.1110342351661697,"vramAllocatedGb":11.72648704,"vramAllocatedPct":6.123395751400361,"vramReservedGb":11.792285696,"vramReservedPct":6.157754823236956,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":24130.59797054676,"meanTps":23960.92496466576,"stepMs":169.7429962158203,"jitter":0.02720164735200319,"achievedTflops":93.68577638597063,"nominalPeakTflops":2250.0,"mfuNominalPct":4.163812283820916,"configuredPeakTflops":2250.0,"mfuConfiguredPct":4.163812283820916,"vramAllocatedGb":16.351640576,"vramAllocatedPct":8.53858159651402,"vramReservedGb":16.697524224,"vramReservedPct":8.719196852678754,"warmupSteps":15,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":49084.629219316004,"meanTps":49113.987620152824,"stepMs":166.89542388916016,"jitter":0.02159989818661828,"achievedTflops":190.5684891291123,"nominalPeakTflops":2250.0,"mfuNominalPct":8.469710627960545,"configuredPeakTflops":2250.0,"mfuConfiguredPct":8.469710627960545,"vramAllocatedGb":25.60191488,"vramAllocatedPct":13.36893617578293,"vramReservedGb":26.254245888,"vramReservedPct":13.709573649671606,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":80343.87845882095,"meanTps":80199.7118367867,"stepMs":203.9234390258789,"jitter":0.006959418852869242,"achievedTflops":311.9308788146103,"nominalPeakTflops":2250.0,"mfuNominalPct":13.86359461398268,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.86359461398268,"vramAllocatedGb":44.102463488,"vramAllocatedPct":23.02964533432075,"vramReservedGb":45.296386048,"vramReservedPct":23.653093798167347,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":89985.59917749812,"meanTps":89719.02540490734,"stepMs":364.14715576171875,"jitter":0.0023660967884513175,"achievedTflops":349.36435196469574,"nominalPeakTflops":2250.0,"mfuNominalPct":15.527304531764255,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.527304531764255,"vramAllocatedGb":81.103560704,"vramAllocatedPct":42.351063651396395,"vramReservedGb":83.554729984,"vramReservedPct":43.63102750620786,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":95721.17722031698,"meanTps":95655.39329614639,"stepMs":684.6551818847656,"jitter":0.003807613018375352,"achievedTflops":371.63243179511176,"nominalPeakTflops":2250.0,"mfuNominalPct":16.516996968671634,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.516996968671634,"vramAllocatedGb":155.105755136,"vramAllocatedPct":80.99390028554767,"vramReservedGb":160.121749504,"vramReservedPct":83.61317735439943,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.221983232,"vramAllocatedPct":99.33106820248355,"vramReservedGb":190.446567424,"vramReservedPct":99.44834270101269,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 193.81 MiB is free. Including non-PyTorch memory, this process has 178.15 GiB memory in use. Of the allocated memory 177.16 GiB is allocated by PyTorch, and 174.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11831.675479380716,"meanTps":11793.447494642554,"stepMs":173.09467315673828,"jitter":0.007124207235894607,"achievedTflops":45.9358572292984,"nominalPeakTflops":2250.0,"mfuNominalPct":2.0415936546354843,"configuredPeakTflops":2250.0,"mfuConfiguredPct":2.0415936546354843,"vramAllocatedGb":11.72648704,"vramAllocatedPct":6.123395751400361,"vramReservedGb":11.792285696,"vramReservedPct":6.157754823236956,"warmupSteps":15,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":23348.93602845132,"meanTps":23822.44590196942,"stepMs":175.42555236816406,"jitter":0.027242302962183284,"achievedTflops":90.65101504246972,"nominalPeakTflops":2250.0,"mfuNominalPct":4.028934001887543,"configuredPeakTflops":2250.0,"mfuConfiguredPct":4.028934001887543,"vramAllocatedGb":16.351640576,"vramAllocatedPct":8.53858159651402,"vramReservedGb":16.697524224,"vramReservedPct":8.719196852678754,"warmupSteps":15,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":46524.48202962762,"meanTps":46162.823750892836,"stepMs":176.07933807373047,"jitter":0.02184077193575079,"achievedTflops":180.62885243129523,"nominalPeakTflops":2250.0,"mfuNominalPct":8.027948996946455,"configuredPeakTflops":2250.0,"mfuConfiguredPct":8.027948996946455,"vramAllocatedGb":25.60191488,"vramAllocatedPct":13.36893617578293,"vramReservedGb":26.254245888,"vramReservedPct":13.709573649671606,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":80565.03202629783,"meanTps":79929.24129548023,"stepMs":203.36366271972656,"jitter":0.0033131907664679743,"achievedTflops":312.78949589881546,"nominalPeakTflops":2250.0,"mfuNominalPct":13.901755373280688,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.901755373280688,"vramAllocatedGb":44.102463488,"vramAllocatedPct":23.02964533432075,"vramReservedGb":45.296386048,"vramReservedPct":23.653093798167347,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":89727.16416882724,"meanTps":89691.00293280746,"stepMs":365.19598388671875,"jitter":0.0016816666641597184,"achievedTflops":348.3609916475499,"nominalPeakTflops":2250.0,"mfuNominalPct":15.482710739891106,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.482710739891106,"vramAllocatedGb":81.103560704,"vramAllocatedPct":42.351063651396395,"vramReservedGb":83.554729984,"vramReservedPct":43.63102750620786,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":95876.17891207144,"meanTps":95859.49696686414,"stepMs":683.5483093261719,"jitter":0.000847190127950916,"achievedTflops":372.23421770405946,"nominalPeakTflops":2250.0,"mfuNominalPct":16.54374300906931,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.54374300906931,"vramAllocatedGb":155.105755136,"vramAllocatedPct":80.99390028554767,"vramReservedGb":160.121749504,"vramReservedPct":83.61317735439943,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.221983232,"vramAllocatedPct":99.33106820248355,"vramReservedGb":190.446567424,"vramReservedPct":99.44834270101269,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 193.81 MiB is free. Including non-PyTorch memory, this process has 178.15 GiB memory in use. Of the allocated memory 177.16 GiB is allocated by PyTorch, and 174.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":23020.518829737895,"meanTps":22656.718586181974,"stepMs":88.964111328125,"jitter":0.006787889153176343,"achievedTflops":89.37595255634619,"nominalPeakTflops":2250.0,"mfuNominalPct":3.9722645580598313,"configuredPeakTflops":2250.0,"mfuConfiguredPct":3.9722645580598313,"vramAllocatedGb":11.72648704,"vramAllocatedPct":4.0797907295067075,"vramReservedGb":11.792285696,"vramReservedPct":4.102682900524942,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":44680.7332464804,"meanTps":43931.021515293134,"stepMs":91.67262268066406,"jitter":0.0023011940863439323,"achievedTflops":173.47059483566207,"nominalPeakTflops":2250.0,"mfuNominalPct":7.709804214918315,"configuredPeakTflops":2250.0,"mfuConfiguredPct":7.709804214918315,"vramAllocatedGb":16.351640576,"vramAllocatedPct":5.688939185847641,"vramReservedGb":16.697524224,"vramReservedPct":5.80927641009774,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":64252.50484201288,"meanTps":64256.25360460504,"stepMs":127.49697494506836,"jitter":0.0009334516944793617,"achievedTflops":249.45696779726012,"nominalPeakTflops":2250.0,"mfuNominalPct":11.086976346544894,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.086976346544894,"vramAllocatedGb":25.60191488,"vramAllocatedPct":8.907224698134645,"vramReservedGb":26.254245888,"vramReservedPct":9.134178771415927,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":78473.60237541312,"meanTps":78478.13888094934,"stepMs":208.78358459472656,"jitter":0.0006647018481030364,"achievedTflops":304.66963037211235,"nominalPeakTflops":2250.0,"mfuNominalPct":13.540872460982772,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.540872460982772,"vramAllocatedGb":44.102463488,"vramAllocatedPct":15.343795722708652,"vramReservedGb":45.296386048,"vramReservedPct":15.75917623482807,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":88866.58336524898,"meanTps":88862.95313220026,"stepMs":368.73252868652344,"jitter":0.0003398931234121055,"achievedTflops":345.01983197863086,"nominalPeakTflops":2250.0,"mfuNominalPct":15.334214754605815,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.334214754605815,"vramAllocatedGb":81.103560704,"vramAllocatedPct":28.216937771856667,"vramReservedGb":83.554729984,"vramReservedPct":29.06973005917033,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":96298.76944257812,"meanTps":96256.4373131196,"stepMs":680.5486755371094,"jitter":0.00024991204794894617,"achievedTflops":373.8749031935861,"nominalPeakTflops":2250.0,"mfuNominalPct":16.61666236415938,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.61666236415938,"vramAllocatedGb":155.105755136,"vramAllocatedPct":53.963221870152694,"vramReservedGb":160.121749504,"vramReservedPct":55.70834871436607,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":277.956605952,"vramAllocatedPct":96.70456124668334,"vramReservedGb":279.2357888,"vramReservedPct":97.14960487371441,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.77 GiB is free. Including non-PyTorch memory, this process has 259.90 GiB memory in use. Of the allocated memory 258.87 GiB is allocated by PyTorch, and 219.92 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":23516.780673091198,"meanTps":22872.8728550892,"stepMs":87.08675003051758,"jitter":0.002382042342223047,"achievedTflops":91.30266303994193,"nominalPeakTflops":2250.0,"mfuNominalPct":4.05789613510853,"configuredPeakTflops":2250.0,"mfuConfiguredPct":4.05789613510853,"vramAllocatedGb":11.72648704,"vramAllocatedPct":4.0797907295067075,"vramReservedGb":11.792285696,"vramReservedPct":4.102682900524942,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":44977.12668972865,"meanTps":44968.620668031945,"stepMs":91.06851196289062,"jitter":0.0009588153996487421,"achievedTflops":174.6213267769225,"nominalPeakTflops":2250.0,"mfuNominalPct":7.760947856752112,"configuredPeakTflops":2250.0,"mfuConfiguredPct":7.760947856752112,"vramAllocatedGb":16.351640576,"vramAllocatedPct":5.688939185847641,"vramReservedGb":16.697524224,"vramReservedPct":5.80927641009774,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":64371.56768787079,"meanTps":63875.35614016895,"stepMs":127.26115417480469,"jitter":0.001441413975489998,"achievedTflops":249.91922302883526,"nominalPeakTflops":2250.0,"mfuNominalPct":11.10752102350379,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.10752102350379,"vramAllocatedGb":25.60191488,"vramAllocatedPct":8.907224698134645,"vramReservedGb":26.254245888,"vramReservedPct":9.134178771415927,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":78496.87117678946,"meanTps":78507.58370754348,"stepMs":208.72169494628906,"jitter":0.0006643028163685591,"achievedTflops":304.7599702685863,"nominalPeakTflops":2250.0,"mfuNominalPct":13.544887567492724,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.544887567492724,"vramAllocatedGb":44.102463488,"vramAllocatedPct":15.343795722708652,"vramReservedGb":45.296386048,"vramReservedPct":15.75917623482807,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":88858.4679682128,"meanTps":88756.43435835968,"stepMs":368.7662048339844,"jitter":0.0003191881629936979,"achievedTflops":344.9883243768324,"nominalPeakTflops":2250.0,"mfuNominalPct":15.332814416748105,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.332814416748105,"vramAllocatedGb":81.103560704,"vramAllocatedPct":28.216937771856667,"vramReservedGb":83.554729984,"vramReservedPct":29.06973005917033,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":96282.70809026156,"meanTps":96231.33955371773,"stepMs":680.6622009277344,"jitter":0.0002541246100072279,"achievedTflops":373.8125458386866,"nominalPeakTflops":2250.0,"mfuNominalPct":16.61389092616385,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.61389092616385,"vramAllocatedGb":155.105755136,"vramAllocatedPct":53.963221870152694,"vramReservedGb":160.121749504,"vramReservedPct":55.70834871436607,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":277.956605952,"vramAllocatedPct":96.70456124668334,"vramReservedGb":279.2357888,"vramReservedPct":97.14960487371441,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.77 GiB is free. Including non-PyTorch memory, this process has 259.90 GiB memory in use. Of the allocated memory 258.87 GiB is allocated by PyTorch, and 219.92 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":18481.654217295167,"meanTps":18473.18947489211,"stepMs":110.81259155273438,"jitter":0.0024921930832663424,"achievedTflops":71.75404962437048,"nominalPeakTflops":165.2,"mfuNominalPct":43.434654736301745,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":34.913980840663605,"vramAllocatedGb":11.72648704,"vramAllocatedPct":46.42122183184461,"vramReservedGb":11.792285696,"vramReservedPct":46.681696601142036,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":21355.578388049773,"meanTps":21355.30324524839,"stepMs":191.8000030517578,"jitter":0.0008233452547593219,"achievedTflops":82.91190893395691,"nominalPeakTflops":165.2,"mfuNominalPct":50.18880686074874,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":40.34315575409468,"vramAllocatedGb":16.351640576,"vramAllocatedPct":64.73065052678277,"vramReservedGb":16.697524224,"vramReservedPct":66.09988766464394,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.029817344,"vramAllocatedPct":95.12597231374512,"vramReservedGb":24.154996736,"vramReservedPct":95.62151546362333,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 595.69 MiB is free. Including non-PyTorch memory, this process has 22.94 GiB memory in use. Of the allocated memory 22.38 GiB is allocated by PyTorch, and 99.38 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":18559.100143575502,"meanTps":18554.566635464227,"stepMs":110.35017776489258,"jitter":0.001212827940284234,"achievedTflops":72.05472935640033,"nominalPeakTflops":165.2,"mfuNominalPct":43.616664259322235,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":35.060285146249456,"vramAllocatedGb":11.72648704,"vramAllocatedPct":46.42122183184461,"vramReservedGb":11.792285696,"vramReservedPct":46.681696601142036,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":21367.025307745218,"meanTps":21366.119255392783,"stepMs":191.69725036621094,"jitter":0.0004647334079382342,"achievedTflops":82.95635099710861,"nominalPeakTflops":165.2,"mfuNominalPct":50.21570883602217,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":40.364780308381455,"vramAllocatedGb":16.351640576,"vramAllocatedPct":64.73065052678277,"vramReservedGb":16.697524224,"vramReservedPct":66.09988766464394,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.029817344,"vramAllocatedPct":95.12597231374512,"vramReservedGb":24.154996736,"vramReservedPct":95.62151546362333,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 595.69 MiB is free. Including non-PyTorch memory, this process has 22.94 GiB memory in use. Of the allocated memory 22.38 GiB is allocated by PyTorch, and 99.38 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":25184.243913399787,"meanTps":25183.458913850198,"stepMs":81.32068634033203,"jitter":0.0008715723109355451,"achievedTflops":97.77650129517507,"nominalPeakTflops":209.5,"mfuNominalPct":46.67136100008356,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":36.35532877902949,"vramAllocatedGb":11.72648704,"vramAllocatedPct":34.82888279178816,"vramReservedGb":11.792285696,"vramReservedPct":35.02431162481071,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":31005.21965838586,"meanTps":31003.446881914773,"stepMs":132.1067886352539,"jitter":0.0003480562878313483,"achievedTflops":120.37613320892028,"nominalPeakTflops":209.5,"mfuNominalPct":57.45877480139392,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":44.75834011228323,"vramAllocatedGb":16.351640576,"vramAllocatedPct":48.56606851926828,"vramReservedGb":16.697524224,"vramReservedPct":49.59337882922689,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":33165.10476237543,"meanTps":33167.68507963269,"stepMs":247.00660705566406,"jitter":0.00020785749166426002,"achievedTflops":128.7617734288079,"nominalPeakTflops":209.5,"mfuNominalPct":61.46146703045724,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":47.87629486806168,"vramAllocatedGb":25.60191488,"vramAllocatedPct":76.04034264986979,"vramReservedGb":26.254245888,"vramReservedPct":77.9778334040557,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.949152256,"vramAllocatedPct":97.86239971941387,"vramReservedGb":33.004978176,"vramReservedPct":98.02820849213424,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 21.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.69 GiB is allocated by PyTorch, and 53.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":25161.787105781215,"meanTps":25146.72635795921,"stepMs":81.39326477050781,"jitter":0.0008813569597927632,"achievedTflops":97.68931392172232,"nominalPeakTflops":209.5,"mfuNominalPct":46.62974411538058,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":36.32291070735309,"vramAllocatedGb":11.72648704,"vramAllocatedPct":34.82888279178816,"vramReservedGb":11.792285696,"vramReservedPct":35.02431162481071,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":30924.97777992826,"meanTps":30922.27036709586,"stepMs":132.44956970214844,"jitter":0.00048721333515356474,"achievedTflops":120.06459833974114,"nominalPeakTflops":209.5,"mfuNominalPct":57.31007080655901,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":44.64250499397655,"vramAllocatedGb":16.351640576,"vramAllocatedPct":48.56606851926828,"vramReservedGb":16.697524224,"vramReservedPct":49.59337882922689,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":33151.56690704038,"meanTps":33151.35073509431,"stepMs":247.10747528076172,"jitter":0.0001912936514520682,"achievedTflops":128.7092134181023,"nominalPeakTflops":209.5,"mfuNominalPct":61.4363787198579,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":47.85675196721617,"vramAllocatedGb":25.60191488,"vramAllocatedPct":76.04034264986979,"vramReservedGb":26.254245888,"vramReservedPct":77.9778334040557,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.949152256,"vramAllocatedPct":97.86239971941387,"vramReservedGb":33.004978176,"vramReservedPct":98.02820849213424,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 21.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.69 GiB is allocated by PyTorch, and 53.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":33148.40137510649,"meanTps":33257.80501350074,"stepMs":61.78276824951172,"jitter":0.005831862464406295,"achievedTflops":128.69692340700257,"nominalPeakTflops":989.5,"mfuNominalPct":13.006258050227649,"configuredPeakTflops":989.5,"mfuConfiguredPct":13.006258050227649,"vramAllocatedGb":11.776556544,"vramAllocatedPct":13.85192159710745,"vramReservedGb":11.834228736,"vramReservedPct":13.919757273770026,"warmupSteps":14,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":45999.97746573003,"meanTps":45996.96075471728,"stepMs":89.04352188110352,"jitter":0.0013653990654275674,"achievedTflops":178.59249107188302,"nominalPeakTflops":989.5,"mfuNominalPct":18.048761098724913,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.048761098724913,"vramAllocatedGb":16.40171008,"vramAllocatedPct":19.292159065155587,"vramReservedGb":16.739467264,"vramReservedPct":19.689438695593186,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":54070.72297953984,"meanTps":54072.78916071604,"stepMs":151.50527954101562,"jitter":0.0005603957881166879,"achievedTflops":209.92673568520578,"nominalPeakTflops":989.5,"mfuNominalPct":21.21543564276966,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.21543564276966,"vramAllocatedGb":25.651984384,"vramAllocatedPct":30.172595458595936,"vramReservedGb":26.296188928,"vramReservedPct":30.93032721173177,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":59425.68061255268,"meanTps":59417.908353618346,"stepMs":275.7057189941406,"jitter":0.0005069278999814596,"achievedTflops":230.71707681040832,"nominalPeakTflops":989.5,"mfuNominalPct":23.316531259263094,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.316531259263094,"vramAllocatedGb":44.152532992,"vramAllocatedPct":51.93346824547663,"vramReservedGb":45.338329088,"vramReservedPct":53.328235424709234,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":63117.265041509876,"meanTps":63117.60341290399,"stepMs":519.16064453125,"jitter":0.00020968526890893032,"achievedTflops":245.04945903083032,"nominalPeakTflops":989.5,"mfuNominalPct":24.76497817390908,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.76497817390908,"vramAllocatedGb":81.153630208,"vramAllocatedPct":95.45521381923803,"vramReservedGb":83.722502144,"vramReservedPct":98.47679423771876,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.881759232,"vramAllocatedPct":98.66411696559067,"vramReservedGb":84.079017984,"vramReservedPct":98.89613833421724,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 142.19 MiB is free. Including non-PyTorch memory, this process has 79.03 GiB memory in use. Of the allocated memory 78.12 GiB is allocated by PyTorch, and 188.12 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":32249.302718963543,"meanTps":32228.69380322037,"stepMs":63.5052490234375,"jitter":0.0030723335848222436,"achievedTflops":125.20622020308083,"nominalPeakTflops":989.5,"mfuNominalPct":12.653483598088007,"configuredPeakTflops":989.5,"mfuConfiguredPct":12.653483598088007,"vramAllocatedGb":11.776556544,"vramAllocatedPct":13.85192159710745,"vramReservedGb":11.834228736,"vramReservedPct":13.919757273770026,"warmupSteps":16,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":45907.029508899854,"meanTps":45887.267729649386,"stepMs":89.22380828857422,"jitter":0.001107632395171061,"achievedTflops":178.23162552227032,"nominalPeakTflops":989.5,"mfuNominalPct":18.01229161417588,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.01229161417588,"vramAllocatedGb":16.40171008,"vramAllocatedPct":19.292159065155587,"vramReservedGb":16.739467264,"vramReservedPct":19.689438695593186,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":53979.06395694311,"meanTps":53979.721206429385,"stepMs":151.76254272460938,"jitter":0.00045122368893344095,"achievedTflops":209.5708743549049,"nominalPeakTflops":989.5,"mfuNominalPct":21.17947189033905,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.17947189033905,"vramAllocatedGb":25.651984384,"vramAllocatedPct":30.172595458595936,"vramReservedGb":26.296188928,"vramReservedPct":30.93032721173177,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":59348.27902498395,"meanTps":59351.73635704344,"stepMs":276.06529235839844,"jitter":0.0005619095395450473,"achievedTflops":230.41656922109232,"nominalPeakTflops":989.5,"mfuNominalPct":23.286161619109887,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.286161619109887,"vramAllocatedGb":44.152532992,"vramAllocatedPct":51.93346824547663,"vramReservedGb":45.338329088,"vramReservedPct":53.328235424709234,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":63173.048816032446,"meanTps":63149.54983541058,"stepMs":518.7022094726562,"jitter":0.0002852355915225153,"achievedTflops":245.2660365989563,"nominalPeakTflops":989.5,"mfuNominalPct":24.786865750273503,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.786865750273503,"vramAllocatedGb":81.153630208,"vramAllocatedPct":95.45521381923803,"vramReservedGb":83.722502144,"vramReservedPct":98.47679423771876,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.881759232,"vramAllocatedPct":98.66411696559067,"vramReservedGb":84.079017984,"vramReservedPct":98.89613833421724,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 142.19 MiB is free. Including non-PyTorch memory, this process has 79.03 GiB memory in use. Of the allocated memory 78.12 GiB is allocated by PyTorch, and 188.12 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":32479.8081137311,"meanTps":32717.09181374682,"stepMs":63.0545597076416,"jitter":0.007746855720402486,"achievedTflops":126.10114526446189,"nominalPeakTflops":989.5,"mfuNominalPct":12.743925746787456,"configuredPeakTflops":989.5,"mfuConfiguredPct":12.743925746787456,"vramAllocatedGb":11.776556544,"vramAllocatedPct":7.844722208458607,"vramReservedGb":11.834228736,"vramReservedPct":7.883139408231947,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":48852.8745409501,"meanTps":48851.46694713241,"stepMs":83.84358215332031,"jitter":0.0014398273140673872,"achievedTflops":189.66871378992292,"nominalPeakTflops":989.5,"mfuNominalPct":19.168136815555627,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.168136815555627,"vramAllocatedGb":16.40171008,"vramAllocatedPct":10.925677539146998,"vramReservedGb":16.739467264,"vramReservedPct":11.150667863992096,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":58884.01474516312,"meanTps":58890.571397174375,"stepMs":139.1209487915039,"jitter":0.0007189872080735904,"achievedTflops":228.61408759355962,"nominalPeakTflops":989.5,"mfuNominalPct":23.104000767413808,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.104000767413808,"vramAllocatedGb":25.651984384,"vramAllocatedPct":17.087566372763145,"vramReservedGb":26.296188928,"vramReservedPct":17.516690597168243,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":65384.60056859917,"meanTps":65383.49710341899,"stepMs":250.5788803100586,"jitter":0.0013172998983828794,"achievedTflops":253.85226986221235,"nominalPeakTflops":989.5,"mfuNominalPct":25.654600289258447,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.654600289258447,"vramAllocatedGb":44.152532992,"vramAllocatedPct":29.41134403999544,"vramReservedGb":45.338329088,"vramReservedPct":30.201238856382503,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":69141.03571516651,"meanTps":69160.35084541823,"stepMs":473.9298400878906,"jitter":0.0014645342131767494,"achievedTflops":268.43643158001396,"nominalPeakTflops":989.5,"mfuNominalPct":27.12849232743951,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.12849232743951,"vramAllocatedGb":81.153630208,"vramAllocatedPct":54.05889937446003,"vramReservedGb":83.722502144,"vramReservedPct":55.77010304012685,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.579055104,"vramAllocatedPct":94.97624164215046,"vramReservedGb":143.428419584,"vramReservedPct":95.54202913482179,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.09 GiB is free. Including non-PyTorch memory, this process has 133.72 GiB memory in use. Of the allocated memory 132.79 GiB is allocated by PyTorch, and 210.02 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":31154.597266858545,"meanTps":31271.467596395418,"stepMs":65.73668670654297,"jitter":0.02080744538307259,"achievedTflops":120.95608391057817,"nominalPeakTflops":989.5,"mfuNominalPct":12.223959970750698,"configuredPeakTflops":989.5,"mfuConfiguredPct":12.223959970750698,"vramAllocatedGb":11.776556544,"vramAllocatedPct":7.844722208458607,"vramReservedGb":11.834228736,"vramReservedPct":7.883139408231947,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":49146.85103944036,"meanTps":49137.5966469521,"stepMs":83.3420639038086,"jitter":0.0013454787070190063,"achievedTflops":190.81006207038848,"nominalPeakTflops":989.5,"mfuNominalPct":19.28348277618883,"configuredPeakTflops":989.5,"mfuConfiguredPct":19.28348277618883,"vramAllocatedGb":16.40171008,"vramAllocatedPct":10.925677539146998,"vramReservedGb":16.739467264,"vramReservedPct":11.150667863992096,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":59021.20651533121,"meanTps":59027.32778011734,"stepMs":138.79756927490234,"jitter":0.000497691391633079,"achievedTflops":229.14672741946245,"nominalPeakTflops":989.5,"mfuNominalPct":23.157829956489383,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.157829956489383,"vramAllocatedGb":25.651984384,"vramAllocatedPct":17.087566372763145,"vramReservedGb":26.296188928,"vramReservedPct":17.516690597168243,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":65509.310873408205,"meanTps":65510.51791194746,"stepMs":250.1018524169922,"jitter":0.0007675915482431322,"achievedTflops":254.33645105588297,"nominalPeakTflops":989.5,"mfuNominalPct":25.70353219362132,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.70353219362132,"vramAllocatedGb":44.152532992,"vramAllocatedPct":29.41134403999544,"vramReservedGb":45.338329088,"vramReservedPct":30.201238856382503,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":69229.57352581633,"meanTps":69194.7300557214,"stepMs":473.32373046875,"jitter":0.0009792792422787242,"achievedTflops":268.78017496923155,"nominalPeakTflops":989.5,"mfuNominalPct":27.163231426905664,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.163231426905664,"vramAllocatedGb":81.153630208,"vramAllocatedPct":54.05889937446003,"vramReservedGb":83.722502144,"vramReservedPct":55.77010304012685,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.579055104,"vramAllocatedPct":94.97624164215046,"vramReservedGb":143.428419584,"vramReservedPct":95.54202913482179,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.09 GiB is free. Including non-PyTorch memory, this process has 133.72 GiB memory in use. Of the allocated memory 132.79 GiB is allocated by PyTorch, and 210.02 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":16862.15383049806,"meanTps":16846.30237015182,"stepMs":121.4554214477539,"jitter":0.006233718515233452,"achievedTflops":65.46642462313109,"nominalPeakTflops":364.2,"mfuNominalPct":17.975404893775696,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.501896701741746,"vramAllocatedGb":11.72648704,"vramAllocatedPct":23.04877146991121,"vramReservedGb":11.792285696,"vramReservedPct":23.178100754973148,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":19809.66679250244,"meanTps":19800.544420921084,"stepMs":206.76773834228516,"jitter":0.007168613677321676,"achievedTflops":76.90998854103077,"nominalPeakTflops":364.2,"mfuNominalPct":21.117514700996917,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.03683553172215,"vramAllocatedGb":16.351640576,"vramAllocatedPct":32.13965320634945,"vramReservedGb":16.697524224,"vramReservedPct":32.819498170210956,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":19976.28360343459,"meanTps":19945.07218109249,"stepMs":410.08628845214844,"jitter":0.0037970523617960874,"achievedTflops":77.55686953876591,"nominalPeakTflops":364.2,"mfuNominalPct":21.295131669073562,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.180130380363696,"vramAllocatedGb":25.60191488,"vramAllocatedPct":50.32135227271264,"vramReservedGb":26.254245888,"vramReservedPct":51.60352896167684,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19292.350026821736,"meanTps":19287.65726718636,"stepMs":849.24853515625,"jitter":0.0017309677069743989,"achievedTflops":74.90153342982998,"nominalPeakTflops":364.2,"mfuNominalPct":20.566044324500268,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.591929479186206,"vramAllocatedGb":44.102463488,"vramAllocatedPct":86.684750405439,"vramReservedGb":45.296386048,"vramReservedPct":89.03144197166371,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":16864.95063538351,"meanTps":16840.716394646235,"stepMs":121.4352798461914,"jitter":0.005585470816463018,"achievedTflops":65.47728307087503,"nominalPeakTflops":364.2,"mfuNominalPct":17.97838634565487,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.504302027653926,"vramAllocatedGb":11.72648704,"vramAllocatedPct":23.04877146991121,"vramReservedGb":11.792285696,"vramReservedPct":23.178100754973148,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":19951.403937332474,"meanTps":19930.573400141435,"stepMs":205.29883575439453,"jitter":0.004319325640942454,"achievedTflops":77.46027554479006,"nominalPeakTflops":364.2,"mfuNominalPct":21.268609430200456,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.158733211804233,"vramAllocatedGb":16.351640576,"vramAllocatedPct":32.13965320634945,"vramReservedGb":16.697524224,"vramReservedPct":32.819498170210956,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":19926.507030700508,"meanTps":19909.213665235027,"stepMs":411.1106872558594,"jitter":0.004147310265732085,"achievedTflops":77.36361461536463,"nominalPeakTflops":364.2,"mfuNominalPct":21.242068812565794,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.137321215934723,"vramAllocatedGb":25.60191488,"vramAllocatedPct":50.32135227271264,"vramReservedGb":26.254245888,"vramReservedPct":51.60352896167684,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19280.689399025447,"meanTps":19266.461557869003,"stepMs":849.7621459960938,"jitter":0.001101953061170107,"achievedTflops":74.85626165622632,"nominalPeakTflops":364.2,"mfuNominalPct":20.55361385398856,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.581901031961785,"vramAllocatedGb":44.102463488,"vramAllocatedPct":86.684750405439,"vramReservedGb":45.296386048,"vramReservedPct":89.03144197166371,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":10129.473572792294,"meanTps":10127.178409682305,"stepMs":202.1822738647461,"jitter":0.0017695239870590801,"achievedTflops":39.32714793087721,"nominalPeakTflops":154.8,"mfuNominalPct":25.405134322272097,"configuredPeakTflops":180.6,"mfuConfiguredPct":21.77582941909037,"vramAllocatedGb":11.72648704,"vramAllocatedPct":22.977824031803635,"vramReservedGb":11.792285696,"vramReservedPct":23.10675522269993,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":12227.414816375684,"meanTps":12225.774156260763,"stepMs":334.9849548339844,"jitter":0.0022860739030464926,"achievedTflops":47.47229437346266,"nominalPeakTflops":154.8,"mfuNominalPct":30.666856830402235,"configuredPeakTflops":180.6,"mfuConfiguredPct":26.28587728320192,"vramAllocatedGb":16.351640576,"vramAllocatedPct":32.0407227249729,"vramReservedGb":16.697524224,"vramReservedPct":32.718475028123216,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":13878.915118274204,"meanTps":13874.25642205876,"stepMs":590.2478637695312,"jitter":0.001375494624388651,"achievedTflops":53.88415736060777,"nominalPeakTflops":154.8,"mfuNominalPct":34.80888718385515,"configuredPeakTflops":180.6,"mfuConfiguredPct":29.836189014732984,"vramAllocatedGb":25.60191488,"vramAllocatedPct":50.16645590305066,"vramReservedGb":26.254245888,"vramReservedPct":51.44468586750497,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":14738.831867953995,"meanTps":14734.032920834903,"stepMs":1111.621337890625,"jitter":0.0007359759979847144,"achievedTflops":57.22273887522186,"nominalPeakTflops":154.8,"mfuNominalPct":36.96559358864461,"configuredPeakTflops":180.6,"mfuConfiguredPct":31.684794504552524,"vramAllocatedGb":44.102463488,"vramAllocatedPct":86.41792225920618,"vramReservedGb":45.296386048,"vramReservedPct":88.75739037081554,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.554755584,"vramAllocatedPct":99.06106354081591,"vramReservedGb":50.671386624,"vramReservedPct":99.28959980275222,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 8.25 MiB is free. Process 1348507 has 47.51 GiB memory in use. Of the allocated memory 47.08 GiB is allocated by PyTorch, and 111.23 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":10121.147633297118,"meanTps":10123.46346669295,"stepMs":202.34859466552734,"jitter":0.002128123138943607,"achievedTflops":39.29482290906465,"nominalPeakTflops":154.8,"mfuNominalPct":25.38425252523556,"configuredPeakTflops":180.6,"mfuConfiguredPct":21.757930735916197,"vramAllocatedGb":11.72648704,"vramAllocatedPct":22.977824031803635,"vramReservedGb":11.792285696,"vramReservedPct":23.10675522269993,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":12269.510550042547,"meanTps":12269.036853379928,"stepMs":333.8356475830078,"jitter":0.0006566517902503458,"achievedTflops":47.635728843504836,"nominalPeakTflops":154.8,"mfuNominalPct":30.772434653426895,"configuredPeakTflops":180.6,"mfuConfiguredPct":26.3763725600802,"vramAllocatedGb":16.351640576,"vramAllocatedPct":32.0407227249729,"vramReservedGb":16.697524224,"vramReservedPct":32.718475028123216,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":13854.568745764567,"meanTps":13848.03976734879,"stepMs":591.2850952148438,"jitter":0.0009278861515450398,"achievedTflops":53.789633850932155,"nominalPeakTflops":154.8,"mfuNominalPct":34.74782548509829,"configuredPeakTflops":180.6,"mfuConfiguredPct":29.78385041579854,"vramAllocatedGb":25.60191488,"vramAllocatedPct":50.16645590305066,"vramReservedGb":26.254245888,"vramReservedPct":51.44468586750497,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":14716.83112819271,"meanTps":14721.073033689347,"stepMs":1113.2831420898438,"jitter":0.0007262614301477887,"achievedTflops":57.13732216121761,"nominalPeakTflops":154.8,"mfuNominalPct":36.91041483282791,"configuredPeakTflops":180.6,"mfuConfiguredPct":31.637498428138212,"vramAllocatedGb":44.102463488,"vramAllocatedPct":86.41792225920618,"vramReservedGb":45.296386048,"vramReservedPct":88.75739037081554,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.554755584,"vramAllocatedPct":99.06106354081591,"vramReservedGb":50.671386624,"vramReservedPct":99.28959980275222,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 8.25 MiB is free. Process 1380474 has 47.51 GiB memory in use. Of the allocated memory 47.08 GiB is allocated by PyTorch, and 111.23 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":25951.428899576596,"meanTps":25947.01838275975,"stepMs":78.91665649414062,"jitter":0.0019946900277307174,"achievedTflops":100.75505661938885,"nominalPeakTflops":468.0,"mfuNominalPct":21.528858252006167,"configuredPeakTflops":468.0,"mfuConfiguredPct":21.528858252006167,"vramAllocatedGb":11.72648704,"vramAllocatedPct":11.499492749513335,"vramReservedGb":11.792285696,"vramReservedPct":11.564017714664342,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":34225.58720876712,"meanTps":34221.67629879931,"stepMs":119.67654418945312,"jitter":0.0022116728517036963,"achievedTflops":132.87904070312769,"nominalPeakTflops":468.0,"mfuNominalPct":28.39295741519822,"configuredPeakTflops":468.0,"mfuConfiguredPct":28.39295741519822,"vramAllocatedGb":16.351640576,"vramAllocatedPct":16.03511534229778,"vramReservedGb":16.697524224,"vramReservedPct":16.374303582457316,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":41180.25460502303,"meanTps":41174.47010335103,"stepMs":198.93029022216797,"jitter":0.0017115108192618516,"achievedTflops":159.8801707753995,"nominalPeakTflops":468.0,"mfuNominalPct":34.162429652863146,"configuredPeakTflops":468.0,"mfuConfiguredPct":34.162429652863146,"vramAllocatedGb":25.60191488,"vramAllocatedPct":25.106328394169918,"vramReservedGb":26.254245888,"vramReservedPct":25.746031970457565,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":39892.005027749816,"meanTps":39883.697324415174,"stepMs":410.7088623046875,"jitter":0.0005911404188463935,"achievedTflops":154.878609605094,"nominalPeakTflops":468.0,"mfuNominalPct":33.093720001088464,"configuredPeakTflops":468.0,"mfuConfiguredPct":33.093720001088464,"vramAllocatedGb":44.102463488,"vramAllocatedPct":43.2487544979142,"vramReservedGb":45.296386048,"vramReservedPct":44.419565822343074,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":39087.62487476393,"meanTps":39086.66158563499,"stepMs":838.3215942382812,"jitter":0.00022214614117840466,"achievedTflops":151.75564600369756,"nominalPeakTflops":468.0,"mfuNominalPct":32.426420086260165,"configuredPeakTflops":468.0,"mfuConfiguredPct":32.426420086260165,"vramAllocatedGb":81.103560704,"vramAllocatedPct":79.53360670540276,"vramReservedGb":83.554729984,"vramReservedPct":81.93732772321833,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":26077.515960897053,"meanTps":26076.72989085464,"stepMs":78.53508758544922,"jitter":0.0005820937285154847,"achievedTflops":101.24458299774263,"nominalPeakTflops":468.0,"mfuNominalPct":21.63345790550056,"configuredPeakTflops":468.0,"mfuConfiguredPct":21.63345790550056,"vramAllocatedGb":11.72648704,"vramAllocatedPct":11.499492749513335,"vramReservedGb":11.792285696,"vramReservedPct":11.564017714664342,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":34445.25629470799,"meanTps":34422.347164993334,"stepMs":118.91332626342773,"jitter":0.00042928812541472965,"achievedTflops":133.73189436591244,"nominalPeakTflops":468.0,"mfuNominalPct":28.575191103827443,"configuredPeakTflops":468.0,"mfuConfiguredPct":28.575191103827443,"vramAllocatedGb":16.351640576,"vramAllocatedPct":16.03511534229778,"vramReservedGb":16.697524224,"vramReservedPct":16.374303582457316,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":41386.49351183963,"meanTps":41379.2743800824,"stepMs":197.93897247314453,"jitter":0.0002900138940190116,"achievedTflops":160.6808824746989,"nominalPeakTflops":468.0,"mfuNominalPct":34.33352189630318,"configuredPeakTflops":468.0,"mfuConfiguredPct":34.33352189630318,"vramAllocatedGb":25.60191488,"vramAllocatedPct":25.106328394169918,"vramReservedGb":26.254245888,"vramReservedPct":25.746031970457565,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":39891.757521803724,"meanTps":39888.68040494175,"stepMs":410.71141052246094,"jitter":0.0002232461099992521,"achievedTflops":154.87764867628704,"nominalPeakTflops":468.0,"mfuNominalPct":33.093514674420305,"configuredPeakTflops":468.0,"mfuConfiguredPct":33.093514674420305,"vramAllocatedGb":44.102463488,"vramAllocatedPct":43.2487544979142,"vramReservedGb":45.296386048,"vramReservedPct":44.419565822343074,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":39109.54434584414,"meanTps":39109.113436178006,"stepMs":837.8517456054688,"jitter":0.00014677643477483622,"achievedTflops":151.8407471963255,"nominalPeakTflops":468.0,"mfuNominalPct":32.44460410177895,"configuredPeakTflops":468.0,"mfuConfiguredPct":32.44460410177895,"vramAllocatedGb":81.103560704,"vramAllocatedPct":79.53360670540276,"vramReservedGb":83.554729984,"vramReservedPct":81.93732772321833,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25297.929392004902,"meanTps":25310.79032559846,"stepMs":647.6419372558594,"jitter":0.0016886968108604256,"achievedTflops":119.70458848567809,"nominalPeakTflops":312.0,"mfuNominalPct":38.36685528387118,"configuredPeakTflops":312.0,"mfuConfiguredPct":38.36685528387118,"vramAllocatedGb":44.110821376,"vramAllocatedPct":51.91081615898146,"vramReservedGb":45.300580352,"vramReservedPct":53.31095693056631,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":26215.5712523289,"meanTps":26219.656733913547,"stepMs":1249.9441528320312,"jitter":0.0010988236437321295,"achievedTflops":124.04668066109669,"nominalPeakTflops":312.0,"mfuNominalPct":39.75855149394125,"configuredPeakTflops":312.0,"mfuConfiguredPct":39.75855149394125,"vramAllocatedGb":81.113198592,"vramAllocatedPct":95.45622159888447,"vramReservedGb":83.626033152,"vramReservedPct":98.41339375784742,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.17457408,"vramAllocatedPct":99.0589316639853,"vramReservedGb":84.360036352,"vramReservedPct":99.27718871760383,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 60.75 MiB is free. Process 764857 has 79.07 GiB memory in use. Of the allocated memory 78.39 GiB is allocated by PyTorch, and 176.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25353.32650691049,"meanTps":25351.438997119232,"stepMs":646.2268371582031,"jitter":0.0017881747737403509,"achievedTflops":119.9667162171739,"nominalPeakTflops":312.0,"mfuNominalPct":38.45087058242753,"configuredPeakTflops":312.0,"mfuConfiguredPct":38.45087058242753,"vramAllocatedGb":44.110821376,"vramAllocatedPct":51.91081615898146,"vramReservedGb":45.300580352,"vramReservedPct":53.31095693056631,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":26296.7481126322,"meanTps":26289.440435007305,"stepMs":1246.0856323242188,"jitter":0.0009352728150119708,"achievedTflops":124.43079283512454,"nominalPeakTflops":312.0,"mfuNominalPct":39.88166437023222,"configuredPeakTflops":312.0,"mfuConfiguredPct":39.88166437023222,"vramAllocatedGb":81.113198592,"vramAllocatedPct":95.45622159888447,"vramReservedGb":83.626033152,"vramReservedPct":98.41339375784742,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.17457408,"vramAllocatedPct":99.0589316639853,"vramReservedGb":84.360036352,"vramReservedPct":99.27718871760383,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 60.75 MiB is free. Process 852916 has 79.07 GiB memory in use. Of the allocated memory 78.39 GiB is allocated by PyTorch, and 176.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":79079.90312665721,"meanTps":79180.81132171441,"stepMs":414.36570739746094,"jitter":0.0005074145623979433,"achievedTflops":374.18980480890565,"nominalPeakTflops":2250.0,"mfuNominalPct":16.63065799150692,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.63065799150692,"vramAllocatedGb":81.113198592,"vramAllocatedPct":42.35609641203735,"vramReservedGb":83.485523968,"vramReservedPct":43.594889162055836,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":83314.84284528387,"meanTps":83295.40398293456,"stepMs":786.6065368652344,"jitter":0.0002621172109679404,"achievedTflops":394.22866682106974,"nominalPeakTflops":2250.0,"mfuNominalPct":17.521274080936433,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.521274080936433,"vramAllocatedGb":155.117953024,"vramAllocatedPct":81.00026983981405,"vramReservedGb":160.249675776,"vramReservedPct":83.67997853601378,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.22906112,"vramAllocatedPct":99.3347641694991,"vramReservedGb":190.496899072,"vramReservedPct":99.47462513312325,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 227.81 MiB is free. Including non-PyTorch memory, this process has 178.12 GiB memory in use. Of the allocated memory 177.16 GiB is allocated by PyTorch, and 135.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":78854.67433689507,"meanTps":78825.78657887367,"stepMs":415.5492401123047,"jitter":0.0011341031052454456,"achievedTflops":373.12406859090015,"nominalPeakTflops":2250.0,"mfuNominalPct":16.58329193737334,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.58329193737334,"vramAllocatedGb":81.113198592,"vramAllocatedPct":42.35609641203735,"vramReservedGb":83.485523968,"vramReservedPct":43.594889162055836,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":83451.3473774374,"meanTps":83436.97818344773,"stepMs":785.3198547363281,"jitter":0.00016817505494132694,"achievedTflops":394.87457813636627,"nominalPeakTflops":2250.0,"mfuNominalPct":17.549981250505166,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.549981250505166,"vramAllocatedGb":155.117953024,"vramAllocatedPct":81.00026983981405,"vramReservedGb":160.249675776,"vramReservedPct":83.67997853601378,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.22906112,"vramAllocatedPct":99.3347641694991,"vramReservedGb":190.496899072,"vramReservedPct":99.47462513312325,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 227.81 MiB is free. Including non-PyTorch memory, this process has 178.12 GiB memory in use. Of the allocated memory 177.16 GiB is allocated by PyTorch, and 135.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":77858.13302642795,"meanTps":77853.6854150681,"stepMs":420.8680419921875,"jitter":0.0005624654448653629,"achievedTflops":368.40864047700296,"nominalPeakTflops":2250.0,"mfuNominalPct":16.373717354533465,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.373717354533465,"vramAllocatedGb":81.113198592,"vramAllocatedPct":28.220290912996063,"vramReservedGb":83.485523968,"vramReservedPct":29.0456524252174,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":83692.83129099835,"meanTps":83690.8107043447,"stepMs":783.0539245605469,"jitter":0.00025787502146898227,"achievedTflops":396.01723025033175,"nominalPeakTflops":2250.0,"mfuNominalPct":17.600765788903633,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.600765788903633,"vramAllocatedGb":155.117953024,"vramAllocatedPct":53.96746566714085,"vramReservedGb":160.249675776,"vramReservedPct":55.75285585591543,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":277.967780352,"vramAllocatedPct":96.70844895946315,"vramReservedGb":279.642636288,"vramReservedPct":97.29115217634678,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.78 GiB is free. Including non-PyTorch memory, this process has 259.89 GiB memory in use. Of the allocated memory 258.88 GiB is allocated by PyTorch, and 197.27 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":77986.93499170226,"meanTps":77986.88656223635,"stepMs":420.1729431152344,"jitter":0.00027604462796843775,"achievedTflops":369.01810483317206,"nominalPeakTflops":2250.0,"mfuNominalPct":16.400804659252092,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.400804659252092,"vramAllocatedGb":81.113198592,"vramAllocatedPct":28.220290912996063,"vramReservedGb":83.485523968,"vramReservedPct":29.0456524252174,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":83753.53675646038,"meanTps":83754.56570187172,"stepMs":782.4863586425781,"jitter":0.0002145795054880532,"achievedTflops":396.3044760027158,"nominalPeakTflops":2250.0,"mfuNominalPct":17.613532266787367,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.613532266787367,"vramAllocatedGb":155.117953024,"vramAllocatedPct":53.96746566714085,"vramReservedGb":160.249675776,"vramReservedPct":55.75285585591543,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":277.967780352,"vramAllocatedPct":96.70844895946315,"vramReservedGb":279.642636288,"vramReservedPct":97.29115217634678,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.78 GiB is free. Including non-PyTorch memory, this process has 259.89 GiB memory in use. Of the allocated memory 258.88 GiB is allocated by PyTorch, and 197.27 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":19804.371053021045,"meanTps":19806.173274197186,"stepMs":206.82302856445312,"jitter":0.0003903290485109065,"achievedTflops":93.71020253810899,"nominalPeakTflops":165.2,"mfuNominalPct":56.725304199823846,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":45.59737371086319,"vramAllocatedGb":16.359005696,"vramAllocatedPct":64.75980656201924,"vramReservedGb":16.45215744,"vramReservedPct":65.12856301546492,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.037151232,"vramAllocatedPct":95.15500471198823,"vramReservedGb":24.209522688,"vramReservedPct":95.8373653856631,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 543.69 MiB is free. Including non-PyTorch memory, this process has 22.99 GiB memory in use. Of the allocated memory 22.39 GiB is allocated by PyTorch, and 144.39 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":19746.08563858417,"meanTps":19746.07902961988,"stepMs":207.4335174560547,"jitter":0.0006301481577327308,"achievedTflops":93.4344079684519,"nominalPeakTflops":165.2,"mfuNominalPct":56.55835833441398,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":45.463177991299276,"vramAllocatedGb":16.359005696,"vramAllocatedPct":64.75980656201924,"vramReservedGb":16.45215744,"vramReservedPct":65.12856301546492,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.037151232,"vramAllocatedPct":95.15500471198823,"vramReservedGb":24.209522688,"vramReservedPct":95.8373653856631,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 543.69 MiB is free. Including non-PyTorch memory, this process has 22.99 GiB memory in use. Of the allocated memory 22.39 GiB is allocated by PyTorch, and 144.39 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":28478.591230081343,"meanTps":28478.984611741234,"stepMs":143.82733917236328,"jitter":0.0003653072923731113,"achievedTflops":134.7548248326642,"nominalPeakTflops":209.5,"mfuNominalPct":64.32211209196383,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":50.104635535714216,"vramAllocatedGb":16.359005696,"vramAllocatedPct":48.587943689585906,"vramReservedGb":16.47312896,"vramReservedPct":48.926901620645225,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":30598.24386542928,"meanTps":30592.445081249538,"stepMs":267.727783203125,"jitter":0.0003073221934264868,"achievedTflops":144.78458428511556,"nominalPeakTflops":209.5,"mfuNominalPct":69.10958677093822,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":53.833907882734074,"vramAllocatedGb":25.609632768,"vramAllocatedPct":76.06326557773686,"vramReservedGb":26.285703168,"vramReservedPct":78.07126478843631,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.720890368,"vramAllocatedPct":97.18443823650209,"vramReservedGb":32.996589568,"vramReservedPct":98.00329345629942,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 108.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 91.88 MiB is free. Including non-PyTorch memory, this process has 31.26 GiB memory in use. Of the allocated memory 30.47 GiB is allocated by PyTorch, and 202.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":28466.37220305656,"meanTps":28464.34210052703,"stepMs":143.88907623291016,"jitter":0.0006228391466327716,"achievedTflops":134.69700691488003,"nominalPeakTflops":209.5,"mfuNominalPct":64.29451404051554,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":50.083137636090896,"vramAllocatedGb":16.359005696,"vramAllocatedPct":48.587943689585906,"vramReservedGb":16.47312896,"vramReservedPct":48.926901620645225,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":30595.0685417386,"meanTps":30586.49235975365,"stepMs":267.7555694580078,"jitter":0.00035823779953405666,"achievedTflops":144.76955930778198,"nominalPeakTflops":209.5,"mfuNominalPct":69.10241494404868,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":53.828321284894876,"vramAllocatedGb":25.609632768,"vramAllocatedPct":76.06326557773686,"vramReservedGb":26.285703168,"vramReservedPct":78.07126478843631,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.720890368,"vramAllocatedPct":97.18443823650209,"vramReservedGb":32.996589568,"vramReservedPct":98.00329345629942,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 108.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 91.88 MiB is free. Including non-PyTorch memory, this process has 31.26 GiB memory in use. Of the allocated memory 30.47 GiB is allocated by PyTorch, and 202.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":53765.66703408697,"meanTps":53760.50143350997,"stepMs":304.7297821044922,"jitter":0.0004102419280657717,"achievedTflops":254.40805637663712,"nominalPeakTflops":989.5,"mfuNominalPct":25.71076870910936,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.71076870910936,"vramAllocatedGb":44.16089088,"vramAllocatedPct":51.94329903165431,"vramReservedGb":45.342523392,"vramReservedPct":53.33316888466804,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":57004.90795422414,"meanTps":56991.64638084765,"stepMs":574.8276977539062,"jitter":0.00047210960485463064,"achievedTflops":269.7354768679574,"nominalPeakTflops":989.5,"mfuNominalPct":27.25977532773698,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.25977532773698,"vramAllocatedGb":81.163268096,"vramAllocatedPct":95.4665501779129,"vramReservedGb":83.626033152,"vramReservedPct":98.36332465866624,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.88883712,"vramAllocatedPct":98.67244217927114,"vramReservedGb":84.087406592,"vramReservedPct":98.90600525413485,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 136.19 MiB is free. Including non-PyTorch memory, this process has 79.04 GiB memory in use. Of the allocated memory 78.13 GiB is allocated by PyTorch, and 189.37 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":53748.84593653878,"meanTps":53719.77780460188,"stepMs":304.8251495361328,"jitter":0.0005468923219332853,"achievedTflops":254.32846240953094,"nominalPeakTflops":989.5,"mfuNominalPct":25.702724851898026,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.702724851898026,"vramAllocatedGb":44.16089088,"vramAllocatedPct":51.94329903165431,"vramReservedGb":45.342523392,"vramReservedPct":53.33316888466804,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":57011.842252885166,"meanTps":57005.60917698388,"stepMs":574.7577819824219,"jitter":0.0002695398340726389,"achievedTflops":269.7682885402011,"nominalPeakTflops":989.5,"mfuNominalPct":27.263091312804562,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.263091312804562,"vramAllocatedGb":81.163268096,"vramAllocatedPct":95.4665501779129,"vramReservedGb":83.626033152,"vramReservedPct":98.36332465866624,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.88883712,"vramAllocatedPct":98.67244217927114,"vramReservedGb":84.087406592,"vramReservedPct":98.90600525413485,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 136.19 MiB is free. Including non-PyTorch memory, this process has 79.04 GiB memory in use. Of the allocated memory 78.13 GiB is allocated by PyTorch, and 189.37 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":58723.77025062594,"meanTps":58750.16355103126,"stepMs":279.00115966796875,"jitter":0.0009790147347880347,"achievedTflops":277.86877903138867,"nominalPeakTflops":989.5,"mfuNominalPct":28.081736132530434,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.081736132530434,"vramAllocatedGb":44.16089088,"vramAllocatedPct":29.41691148319197,"vramReservedGb":45.342523392,"vramReservedPct":30.204032809743563,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":61894.82819882401,"meanTps":61916.64234047301,"stepMs":529.4141845703125,"jitter":0.0008983877881309054,"achievedTflops":292.87357175064,"nominalPeakTflops":989.5,"mfuNominalPct":29.598137620074784,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.598137620074784,"vramAllocatedGb":81.163268096,"vramAllocatedPct":54.0653194645563,"vramReservedGb":83.605061632,"vramReservedPct":55.691872346017156,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.588180992,"vramAllocatedPct":94.98232067348683,"vramReservedGb":143.331950592,"vramReservedPct":95.4777682075174,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.16 GiB is free. Including non-PyTorch memory, this process has 133.64 GiB memory in use. Of the allocated memory 132.80 GiB is allocated by PyTorch, and 129.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":58682.972140419166,"meanTps":58638.73958522317,"stepMs":279.19512939453125,"jitter":0.0007125620500300262,"achievedTflops":277.67573078156477,"nominalPeakTflops":989.5,"mfuNominalPct":28.06222645594389,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.06222645594389,"vramAllocatedGb":44.16089088,"vramAllocatedPct":29.41691148319197,"vramReservedGb":45.342523392,"vramReservedPct":30.204032809743563,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":61780.14145875777,"meanTps":61766.494583950385,"stepMs":530.39697265625,"jitter":0.0014500095051521145,"achievedTflops":292.33089773128995,"nominalPeakTflops":989.5,"mfuNominalPct":29.543294363950476,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.543294363950476,"vramAllocatedGb":81.163268096,"vramAllocatedPct":54.0653194645563,"vramReservedGb":83.605061632,"vramReservedPct":55.691872346017156,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.588180992,"vramAllocatedPct":94.98232067348683,"vramReservedGb":143.331950592,"vramReservedPct":95.4777682075174,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.16 GiB is free. Including non-PyTorch memory, this process has 133.64 GiB memory in use. Of the allocated memory 132.80 GiB is allocated by PyTorch, and 129.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":18770.85502013667,"meanTps":18786.724020987444,"stepMs":218.21062469482422,"jitter":0.0025018512757967702,"achievedTflops":88.81981765748417,"nominalPeakTflops":364.2,"mfuNominalPct":24.387648999858364,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":19.675059821141772,"vramAllocatedGb":16.359005696,"vramAllocatedPct":32.15412957656582,"vramReservedGb":16.45215744,"vramReservedPct":32.3372221986065,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":18603.487845709577,"meanTps":18617.20924502263,"stepMs":440.3475341796875,"jitter":0.0034620318867682356,"achievedTflops":88.02787067912249,"nominalPeakTflops":364.2,"mfuNominalPct":24.17020062578871,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":19.499630456554176,"vramAllocatedGb":25.609632768,"vramAllocatedPct":50.33652201929877,"vramReservedGb":26.306674688,"vramReservedPct":51.70657938295984,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":18000.191176477983,"meanTps":18001.46153365432,"stepMs":910.2125549316406,"jitter":0.0020022113505920786,"achievedTflops":85.17319516769618,"nominalPeakTflops":364.2,"mfuNominalPct":23.386379782453645,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":18.867272578114775,"vramAllocatedGb":44.110821376,"vramAllocatedPct":86.70117809173806,"vramReservedGb":45.43479808,"vramReservedPct":89.30349508385083,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":12944.434481289685,"meanTps":12944.377491749858,"stepMs":632.8588562011719,"jitter":0.0006176332569002107,"achievedTflops":61.25039637640489,"nominalPeakTflops":154.8,"mfuNominalPct":39.567439519641404,"configuredPeakTflops":180.6,"mfuConfiguredPct":33.914948159692635,"vramAllocatedGb":25.609632768,"vramAllocatedPct":50.18157895497203,"vramReservedGb":26.260537344,"vramReservedPct":51.457013853574345,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":13707.734650245906,"meanTps":13708.01969188013,"stepMs":1195.2376098632812,"jitter":0.000718860755828605,"achievedTflops":64.86217547500691,"nominalPeakTflops":154.8,"mfuNominalPct":41.900630151813246,"configuredPeakTflops":180.6,"mfuConfiguredPct":35.914825844411354,"vramAllocatedGb":44.110821376,"vramAllocatedPct":86.43429937872087,"vramReservedGb":45.306871808,"vramReservedPct":88.77793701426451,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.561849856,"vramAllocatedPct":99.07496462927435,"vramReservedGb":50.675580928,"vramReservedPct":99.29781846013181,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 24.25 MiB is free. Process 1415479 has 47.50 GiB memory in use. Of the allocated memory 47.09 GiB is allocated by PyTorch, and 88.46 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":12976.42325164584,"meanTps":12976.348227891192,"stepMs":631.2987670898438,"jitter":0.0005945049979218228,"achievedTflops":61.401760645484416,"nominalPeakTflops":154.8,"mfuNominalPct":39.66522005522249,"configuredPeakTflops":180.6,"mfuConfiguredPct":33.99876004733357,"vramAllocatedGb":25.609632768,"vramAllocatedPct":50.18157895497203,"vramReservedGb":26.260537344,"vramReservedPct":51.457013853574345,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":13746.465694347045,"meanTps":13745.311875263618,"stepMs":1191.8699951171875,"jitter":0.0007177790325261714,"achievedTflops":65.04544279399992,"nominalPeakTflops":154.8,"mfuNominalPct":42.01901989276481,"configuredPeakTflops":180.6,"mfuConfiguredPct":36.01630276522698,"vramAllocatedGb":44.110821376,"vramAllocatedPct":86.43429937872087,"vramReservedGb":45.306871808,"vramReservedPct":88.77793701426451,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.561849856,"vramAllocatedPct":99.07496462927435,"vramReservedGb":50.675580928,"vramReservedPct":99.29781846013181,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 24.25 MiB is free. Process 1449495 has 47.50 GiB memory in use. Of the allocated memory 47.09 GiB is allocated by PyTorch, and 88.46 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":32408.521373609005,"meanTps":32383.410826005507,"stepMs":126.38651275634766,"jitter":0.0012612505887907284,"achievedTflops":153.35044439183312,"nominalPeakTflops":468.0,"mfuNominalPct":32.767188972613916,"configuredPeakTflops":468.0,"mfuConfiguredPct":32.767188972613916,"vramAllocatedGb":16.359005696,"vramAllocatedPct":16.042337892730007,"vramReservedGb":16.47312896,"vramReservedPct":16.15425202715426,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":38721.94585831962,"meanTps":38711.21040484501,"stepMs":211.5596160888672,"jitter":0.0005485122473353509,"achievedTflops":183.22426798296596,"nominalPeakTflops":468.0,"mfuNominalPct":39.15048461174486,"configuredPeakTflops":468.0,"mfuConfiguredPct":39.15048461174486,"vramAllocatedGb":25.609632768,"vramAllocatedPct":25.113896883931158,"vramReservedGb":26.285703168,"vramReservedPct":25.776880319331823,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":37635.9271598446,"meanTps":37633.35769680261,"stepMs":435.32872009277344,"jitter":0.0002630564082633457,"achievedTflops":178.08545130851553,"nominalPeakTflops":468.0,"mfuNominalPct":38.05244686079392,"configuredPeakTflops":468.0,"mfuConfiguredPct":38.05244686079392,"vramAllocatedGb":44.110821376,"vramAllocatedPct":43.25695059893997,"vramReservedGb":45.4557696,"vramReservedPct":44.57586412330599,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":32442.869341425732,"meanTps":32439.699117543438,"stepMs":126.25270462036133,"jitter":0.00034739629752357574,"achievedTflops":153.5129718971127,"nominalPeakTflops":468.0,"mfuNominalPct":32.80191707203263,"configuredPeakTflops":468.0,"mfuConfiguredPct":32.80191707203263,"vramAllocatedGb":16.48345088,"vramAllocatedPct":16.164374141628624,"vramReservedGb":17.664311296,"vramReservedPct":17.32237617119291,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":38568.95961580965,"meanTps":38559.38595192533,"stepMs":212.3987808227539,"jitter":0.0012666217640935284,"achievedTflops":182.50036861081364,"nominalPeakTflops":468.0,"mfuNominalPct":38.995805258720864,"configuredPeakTflops":468.0,"mfuConfiguredPct":38.995805258720864,"vramAllocatedGb":25.674370048,"vramAllocatedPct":25.177381018561064,"vramReservedGb":27.787264,"vramReservedPct":27.249374838929846,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":37535.47759753775,"meanTps":37519.1118394939,"stepMs":436.49371337890625,"jitter":0.00042302171582432976,"achievedTflops":177.61014468032533,"nominalPeakTflops":468.0,"mfuNominalPct":37.950885615454126,"configuredPeakTflops":468.0,"mfuConfiguredPct":37.950885615454126,"vramAllocatedGb":44.183863296,"vramAllocatedPct":43.328578617338444,"vramReservedGb":45.820674048,"vramReservedPct":44.93370497024741,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21630.268997382915,"meanTps":21628.73836327247,"stepMs":757.4570617675781,"jitter":0.001086129427732152,"achievedTflops":139.09316445938464,"nominalPeakTflops":312.0,"mfuNominalPct":44.581142454930976,"configuredPeakTflops":312.0,"mfuConfiguredPct":44.581142454930976,"vramAllocatedGb":44.125617152,"vramAllocatedPct":51.92822823121015,"vramReservedGb":45.315260416,"vramReservedPct":53.32823282976144,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":22359.43288508203,"meanTps":22360.360510523395,"stepMs":1465.51123046875,"jitter":0.0005765922161468424,"achievedTflops":143.7820433892698,"nominalPeakTflops":312.0,"mfuNominalPct":46.083988265791604,"configuredPeakTflops":312.0,"mfuConfiguredPct":46.083988265791604,"vramAllocatedGb":81.128634368,"vramAllocatedPct":95.47438684054653,"vramReservedGb":83.636518912,"vramReservedPct":98.42573368584394,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.188729856,"vramAllocatedPct":99.0755905667806,"vramReservedGb":84.370522112,"vramReservedPct":99.28952864560036,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 50.75 MiB is free. Process 764857 has 79.08 GiB memory in use. Of the allocated memory 78.41 GiB is allocated by PyTorch, and 173.37 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21652.06119916249,"meanTps":21653.711991567525,"stepMs":756.6947021484375,"jitter":0.0011423532181731332,"achievedTflops":139.23329893050126,"nominalPeakTflops":312.0,"mfuNominalPct":44.626057349519634,"configuredPeakTflops":312.0,"mfuConfiguredPct":44.626057349519634,"vramAllocatedGb":44.125617152,"vramAllocatedPct":51.92822823121015,"vramReservedGb":45.315260416,"vramReservedPct":53.32823282976144,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":22385.894012789464,"meanTps":22385.03408219575,"stepMs":1463.7789306640625,"jitter":0.0004638349943880039,"achievedTflops":143.95220132805625,"nominalPeakTflops":312.0,"mfuNominalPct":46.1385260666847,"configuredPeakTflops":312.0,"mfuConfiguredPct":46.1385260666847,"vramAllocatedGb":81.128634368,"vramAllocatedPct":95.47438684054653,"vramReservedGb":83.636518912,"vramReservedPct":98.42573368584394,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.188729856,"vramAllocatedPct":99.0755905667806,"vramReservedGb":84.370522112,"vramReservedPct":99.28952864560036,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 50.75 MiB is free. Process 852916 has 79.08 GiB memory in use. Of the allocated memory 78.41 GiB is allocated by PyTorch, and 173.37 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":62868.3578964342,"meanTps":62849.87005269713,"stepMs":521.2160949707031,"jitter":0.000715031720731329,"achievedTflops":404.27416067910184,"nominalPeakTflops":2250.0,"mfuNominalPct":17.967740474626748,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.967740474626748,"vramAllocatedGb":81.128634368,"vramAllocatedPct":42.36415674288116,"vramReservedGb":83.502301184,"vramReservedPct":43.60364997275936,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":66019.36701716787,"meanTps":65991.69634681815,"stepMs":992.6784057617188,"jitter":0.0002696002156413384,"achievedTflops":424.53668399290143,"nominalPeakTflops":2250.0,"mfuNominalPct":18.868297066351175,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.868297066351175,"vramAllocatedGb":155.1346688,"vramAllocatedPct":81.00899856747056,"vramReservedGb":160.26435584,"vramReservedPct":83.68764424537936,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.243216896,"vramAllocatedPct":99.3421561035302,"vramReservedGb":190.50528768,"vramReservedPct":99.47900553847502,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 219.81 MiB is free. Including non-PyTorch memory, this process has 178.12 GiB memory in use. Of the allocated memory 177.18 GiB is allocated by PyTorch, and 129.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":63003.54142742498,"meanTps":62989.96883487798,"stepMs":520.0977478027344,"jitter":0.0011337052531359278,"achievedTflops":405.14345662315947,"nominalPeakTflops":2250.0,"mfuNominalPct":18.006375849918197,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.006375849918197,"vramAllocatedGb":81.128634368,"vramAllocatedPct":42.36415674288116,"vramReservedGb":83.502301184,"vramReservedPct":43.60364997275936,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":66013.76982410908,"meanTps":65999.55868010278,"stepMs":992.7625732421875,"jitter":0.00022851409573019377,"achievedTflops":424.5006913154765,"nominalPeakTflops":2250.0,"mfuNominalPct":18.866697391798954,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.866697391798954,"vramAllocatedGb":155.1346688,"vramAllocatedPct":81.00899856747056,"vramReservedGb":160.26435584,"vramReservedPct":83.68764424537936,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.243216896,"vramAllocatedPct":99.3421561035302,"vramReservedGb":190.50528768,"vramReservedPct":99.47900553847502,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 219.81 MiB is free. Including non-PyTorch memory, this process has 178.12 GiB memory in use. Of the allocated memory 177.18 GiB is allocated by PyTorch, and 129.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":62418.4999436994,"meanTps":62416.96802980588,"stepMs":524.9725646972656,"jitter":0.0002992382937608172,"achievedTflops":401.3813549441997,"nominalPeakTflops":2250.0,"mfuNominalPct":17.83917133085332,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.83917133085332,"vramAllocatedGb":81.128634368,"vramAllocatedPct":28.22566121150172,"vramReservedGb":83.502301184,"vramReservedPct":29.051489427387807,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":66373.69444112445,"meanTps":66357.19519883451,"stepMs":987.3791198730469,"jitter":0.00016794384844688544,"achievedTflops":426.8151819005708,"nominalPeakTflops":2250.0,"mfuNominalPct":18.96956364002537,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.96956364002537,"vramAllocatedGb":155.1346688,"vramAllocatedPct":53.973281293570885,"vramReservedGb":160.26435584,"vramReservedPct":55.75796323281453,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":277.983984128,"vramAllocatedPct":96.71408645472344,"vramReservedGb":279.651024896,"vramReservedPct":97.29407067743199,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.78 GiB is free. Including non-PyTorch memory, this process has 259.90 GiB memory in use. Of the allocated memory 258.89 GiB is allocated by PyTorch, and 189.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":62480.43810598485,"meanTps":62475.87563333855,"stepMs":524.4521484375,"jitter":0.0005438565914207349,"achievedTflops":401.77964749405777,"nominalPeakTflops":2250.0,"mfuNominalPct":17.856873221958125,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.856873221958125,"vramAllocatedGb":81.128634368,"vramAllocatedPct":28.22566121150172,"vramReservedGb":83.502301184,"vramReservedPct":29.051489427387807,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":66421.15390212482,"meanTps":66386.17312416047,"stepMs":986.6736145019531,"jitter":0.0001275726778995513,"achievedTflops":427.12036934945905,"nominalPeakTflops":2250.0,"mfuNominalPct":18.983127526642626,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.983127526642626,"vramAllocatedGb":155.1346688,"vramAllocatedPct":53.973281293570885,"vramReservedGb":160.26435584,"vramReservedPct":55.75796323281453,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":277.983984128,"vramAllocatedPct":96.71408645472344,"vramReservedGb":279.651024896,"vramReservedPct":97.29407067743199,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.78 GiB is free. Including non-PyTorch memory, this process has 259.90 GiB memory in use. Of the allocated memory 258.89 GiB is allocated by PyTorch, and 189.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.575595008,"vramAllocatedPct":97.28652269213107,"vramReservedGb":24.700256256,"vramReservedPct":97.78001468402114,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 75.69 MiB is free. Including non-PyTorch memory, this process has 23.44 GiB memory in use. Of the allocated memory 22.89 GiB is allocated by PyTorch, and 98.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.575595008,"vramAllocatedPct":97.28652269213107,"vramReservedGb":24.700256256,"vramReservedPct":97.78001468402114,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 75.69 MiB is free. Including non-PyTorch memory, this process has 23.44 GiB memory in use. Of the allocated memory 22.89 GiB is allocated by PyTorch, and 98.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":26722.362358938448,"meanTps":26720.43134617466,"stepMs":306.5597229003906,"jitter":0.0003943623262865232,"achievedTflops":171.83780482733752,"nominalPeakTflops":209.5,"mfuNominalPct":82.02281853333534,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":63.892855731306845,"vramAllocatedGb":25.624043008,"vramAllocatedPct":76.10606548518139,"vramReservedGb":25.748832256,"vramReservedPct":76.47670249500726,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.96153856,"vramAllocatedPct":97.89918832701373,"vramReservedGb":33.011269632,"vramReservedPct":98.04689476901036,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 20.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 17.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.70 GiB is allocated by PyTorch, and 47.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":26711.24081754458,"meanTps":26710.12040200526,"stepMs":306.68736267089844,"jitter":0.0005227699520121908,"achievedTflops":171.76628789953944,"nominalPeakTflops":209.5,"mfuNominalPct":81.98868157495914,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":63.86626425596332,"vramAllocatedGb":25.624043008,"vramAllocatedPct":76.10606548518139,"vramReservedGb":25.748832256,"vramReservedPct":76.47670249500726,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.96153856,"vramAllocatedPct":97.89918832701373,"vramReservedGb":33.011269632,"vramReservedPct":98.04689476901036,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 20.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 17.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.70 GiB is allocated by PyTorch, and 47.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":45232.743203199374,"meanTps":45205.04562795236,"stepMs":362.2154846191406,"jitter":0.0004636764257237525,"achievedTflops":290.8685689518217,"nominalPeakTflops":989.5,"mfuNominalPct":29.39550974753125,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.39550974753125,"vramAllocatedGb":44.175686656,"vramAllocatedPct":51.96070224526388,"vramReservedGb":45.357203456,"vramReservedPct":53.350435994523856,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":47733.647994751416,"meanTps":47723.94278994627,"stepMs":686.4759216308594,"jitter":0.0002388564322308403,"achievedTflops":306.950604802171,"nominalPeakTflops":989.5,"mfuNominalPct":31.020778656106213,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.020778656106213,"vramAllocatedGb":81.178703872,"vramAllocatedPct":95.48470617777106,"vramReservedGb":83.636518912,"vramReservedPct":98.37565830856326,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.902992896,"vramAllocatedPct":98.68909260663212,"vramReservedGb":84.097892352,"vramReservedPct":98.91833890403187,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 126.19 MiB is free. Including non-PyTorch memory, this process has 79.05 GiB memory in use. Of the allocated memory 78.14 GiB is allocated by PyTorch, and 185.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":45227.78376443514,"meanTps":45221.139859363815,"stepMs":362.2552032470703,"jitter":0.0011216566767437112,"achievedTflops":290.83667734512255,"nominalPeakTflops":989.5,"mfuNominalPct":29.392286745338307,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.392286745338307,"vramAllocatedGb":44.175686656,"vramAllocatedPct":51.96070224526388,"vramReservedGb":45.357203456,"vramReservedPct":53.350435994523856,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":47732.39815795772,"meanTps":47718.963045682554,"stepMs":686.493896484375,"jitter":0.00048508203944764336,"achievedTflops":306.942567742867,"nominalPeakTflops":989.5,"mfuNominalPct":31.019966421714706,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.019966421714706,"vramAllocatedGb":81.178703872,"vramAllocatedPct":95.48470617777106,"vramReservedGb":83.636518912,"vramReservedPct":98.37565830856326,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.902992896,"vramAllocatedPct":98.68909260663212,"vramReservedGb":84.097892352,"vramReservedPct":98.91833890403187,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 126.19 MiB is free. Including non-PyTorch memory, this process has 79.05 GiB memory in use. Of the allocated memory 78.14 GiB is allocated by PyTorch, and 185.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":48827.60887502834,"meanTps":48795.05822245686,"stepMs":335.54786682128906,"jitter":0.0006026011182573901,"achievedTflops":313.9853060650587,"nominalPeakTflops":989.5,"mfuNominalPct":31.7317135992985,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.7317135992985,"vramAllocatedGb":44.175686656,"vramAllocatedPct":29.426767399235416,"vramReservedGb":45.357203456,"vramReservedPct":30.213811646507274,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":51213.433699549736,"meanTps":51210.83090397794,"stepMs":639.8321228027344,"jitter":0.0019493845827383825,"achievedTflops":329.3273216788538,"nominalPeakTflops":989.5,"mfuNominalPct":33.28219521767093,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.28219521767093,"vramAllocatedGb":81.178703872,"vramAllocatedPct":54.07560170404962,"vramReservedGb":83.615547392,"vramReservedPct":55.69885722941981,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.603360768,"vramAllocatedPct":94.99243238360019,"vramReservedGb":143.344533504,"vramReservedPct":95.48615006760058,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.15 GiB is free. Including non-PyTorch memory, this process has 133.66 GiB memory in use. Of the allocated memory 132.81 GiB is allocated by PyTorch, and 126.84 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":48684.085734699336,"meanTps":48664.487810467064,"stepMs":336.5370788574219,"jitter":0.0004290105223716356,"achievedTflops":313.062382371233,"nominalPeakTflops":989.5,"mfuNominalPct":31.638441876830015,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.638441876830015,"vramAllocatedGb":44.175686656,"vramAllocatedPct":29.426767399235416,"vramReservedGb":45.357203456,"vramReservedPct":30.213811646507274,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":51069.13174585562,"meanTps":51045.068634450225,"stepMs":641.6400451660156,"jitter":0.002148174350843968,"achievedTflops":328.39938983577713,"nominalPeakTflops":989.5,"mfuNominalPct":33.18841736591987,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.18841736591987,"vramAllocatedGb":81.178703872,"vramAllocatedPct":54.07560170404962,"vramReservedGb":83.615547392,"vramReservedPct":55.69885722941981,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.603360768,"vramAllocatedPct":94.99243238360019,"vramReservedGb":143.344533504,"vramReservedPct":95.48615006760058,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.15 GiB is free. Including non-PyTorch memory, this process has 133.66 GiB memory in use. Of the allocated memory 132.81 GiB is allocated by PyTorch, and 126.84 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":16708.26568671067,"meanTps":16705.951595067465,"stepMs":490.29624938964844,"jitter":0.0017721312315692225,"achievedTflops":107.44228595926965,"nominalPeakTflops":364.2,"mfuNominalPct":29.50090224032665,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":23.800244802582366,"vramAllocatedGb":25.624043008,"vramAllocatedPct":50.36484578987504,"vramReservedGb":25.748832256,"vramReservedPct":50.610122900508685,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":16167.958321328671,"meanTps":16161.764492854883,"stepMs":1013.3623352050781,"jitter":0.0013053954969450118,"achievedTflops":103.96784644856419,"nominalPeakTflops":364.2,"mfuNominalPct":28.54691006275788,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":23.030598939519642,"vramAllocatedGb":44.125617152,"vramAllocatedPct":86.73025964519739,"vramReservedGb":45.321551872,"vramReservedPct":89.08090617387955,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":11520.84688842598,"meanTps":11520.612273140385,"stepMs":711.058837890625,"jitter":0.0015165168324636899,"achievedTflops":74.0846566058476,"nominalPeakTflops":154.8,"mfuNominalPct":47.858305300935136,"configuredPeakTflops":180.6,"mfuConfiguredPct":41.02140454365869,"vramAllocatedGb":25.624043008,"vramAllocatedPct":50.20981554090323,"vramReservedGb":25.748832256,"vramReservedPct":50.45433765326512,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":12110.898802352416,"meanTps":12109.821719781326,"stepMs":1352.8310546875,"jitter":0.000617728149302064,"achievedTflops":77.87897779127873,"nominalPeakTflops":154.8,"mfuNominalPct":50.309417177828635,"configuredPeakTflops":180.6,"mfuConfiguredPct":43.122357580995974,"vramAllocatedGb":44.125617152,"vramAllocatedPct":86.46329141497029,"vramReservedGb":45.321551872,"vramReservedPct":88.80670231509305,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.5760384,"vramAllocatedPct":99.10276680619121,"vramReservedGb":50.669289472,"vramReservedPct":99.28549047406243,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 10.25 MiB is free. Process 1415479 has 47.51 GiB memory in use. Of the allocated memory 47.10 GiB is allocated by PyTorch, and 88.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":11554.603722768323,"meanTps":11553.434748303276,"stepMs":708.9814758300781,"jitter":0.0009187154303532146,"achievedTflops":74.30172949159746,"nominalPeakTflops":154.8,"mfuNominalPct":47.998533263305845,"configuredPeakTflops":180.6,"mfuConfiguredPct":41.141599939976444,"vramAllocatedGb":25.624043008,"vramAllocatedPct":50.20981554090323,"vramReservedGb":25.748832256,"vramReservedPct":50.45433765326512,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":12131.394573147234,"meanTps":12133.135857524527,"stepMs":1350.5454711914062,"jitter":0.0010689003290364311,"achievedTflops":78.01077557974962,"nominalPeakTflops":154.8,"mfuNominalPct":50.3945578680553,"configuredPeakTflops":180.6,"mfuConfiguredPct":43.19533531547598,"vramAllocatedGb":44.125617152,"vramAllocatedPct":86.46329141497029,"vramReservedGb":45.321551872,"vramReservedPct":88.80670231509305,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.5760384,"vramAllocatedPct":99.10276680619121,"vramReservedGb":50.669289472,"vramReservedPct":99.28549047406243,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 10.25 MiB is free. Process 1449495 has 47.51 GiB memory in use. Of the allocated memory 47.10 GiB is allocated by PyTorch, and 88.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":34670.70743182888,"meanTps":34671.70684159728,"stepMs":236.2801513671875,"jitter":0.0002886237818420785,"achievedTflops":222.9495348080073,"nominalPeakTflops":468.0,"mfuNominalPct":47.63878948889045,"configuredPeakTflops":468.0,"mfuConfiguredPct":47.63878948889045,"vramAllocatedGb":25.624043008,"vramAllocatedPct":25.128028179163355,"vramReservedGb":25.748832256,"vramReservedPct":25.250401831877785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":33911.780566953705,"meanTps":33912.423303766394,"stepMs":483.1359405517578,"jitter":0.00018912483947643665,"achievedTflops":218.06926543911965,"nominalPeakTflops":468.0,"mfuNominalPct":46.59599688870078,"configuredPeakTflops":468.0,"mfuConfiguredPct":46.59599688870078,"vramAllocatedGb":44.125617152,"vramAllocatedPct":43.27145996719792,"vramReservedGb":45.321551872,"vramReservedPct":44.44424450144248,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":34536.69444376184,"meanTps":34530.84883460802,"stepMs":237.19699096679688,"jitter":0.0005127364000457512,"achievedTflops":222.0877660250505,"nominalPeakTflops":468.0,"mfuNominalPct":47.454650860053526,"configuredPeakTflops":468.0,"mfuConfiguredPct":47.454650860053526,"vramAllocatedGb":25.673842688,"vramAllocatedPct":25.176863866879092,"vramReservedGb":27.193769984,"vramReservedPct":26.667369323502136,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":33834.53969550415,"meanTps":33834.76742791644,"stepMs":484.2388916015625,"jitter":0.00015893469309042233,"achievedTflops":217.572569016305,"nominalPeakTflops":468.0,"mfuNominalPct":46.48986517442415,"configuredPeakTflops":468.0,"mfuConfiguredPct":46.48986517442415,"vramAllocatedGb":44.183076864,"vramAllocatedPct":43.327807408616586,"vramReservedGb":47.592767488,"vramReservedPct":46.67149529016407,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":16962.967920022715,"meanTps":16959.919435101794,"stepMs":965.8687133789062,"jitter":0.0010190490453808778,"achievedTflops":166.7098991706884,"nominalPeakTflops":312.0,"mfuNominalPct":53.43265999060526,"configuredPeakTflops":312.0,"mfuConfiguredPct":53.43265999060526,"vramAllocatedGb":44.154117632,"vramAllocatedPct":51.96176837241748,"vramReservedGb":44.293947392,"vramReservedPct":52.126323842900376,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":17312.385104544224,"meanTps":17314.236971028982,"stepMs":1892.7490234375,"jitter":0.0011666866962477588,"achievedTflops":170.1439269820202,"nominalPeakTflops":312.0,"mfuNominalPct":54.53330993013468,"configuredPeakTflops":312.0,"mfuConfiguredPct":54.53330993013468,"vramAllocatedGb":81.15758592,"vramAllocatedPct":95.50845781557052,"vramReservedGb":83.667976192,"vramReservedPct":98.4627534698335,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.217041408,"vramAllocatedPct":99.1089083723712,"vramReservedGb":84.401979392,"vramReservedPct":99.32654842958992,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 20.75 MiB is free. Process 925444 has 79.11 GiB memory in use. Of the allocated memory 78.43 GiB is allocated by PyTorch, and 176.37 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":16976.91169861133,"meanTps":16977.653165220436,"stepMs":965.0754089355469,"jitter":0.00038007822071819736,"achievedTflops":166.8469368597017,"nominalPeakTflops":312.0,"mfuNominalPct":53.47658232682747,"configuredPeakTflops":312.0,"mfuConfiguredPct":53.47658232682747,"vramAllocatedGb":44.154117632,"vramAllocatedPct":51.96176837241748,"vramReservedGb":44.293947392,"vramReservedPct":52.126323842900376,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":17262.69476487251,"meanTps":17269.515265373782,"stepMs":1898.197265625,"jitter":0.001016794544217015,"achievedTflops":169.65557662048644,"nominalPeakTflops":312.0,"mfuNominalPct":54.37678737836104,"configuredPeakTflops":312.0,"mfuConfiguredPct":54.37678737836104,"vramAllocatedGb":81.15758592,"vramAllocatedPct":95.50845781557052,"vramReservedGb":83.667976192,"vramReservedPct":98.4627534698335,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.217041408,"vramAllocatedPct":99.1089083723712,"vramReservedGb":84.401979392,"vramReservedPct":99.32654842958992,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 20.75 MiB is free. Process 937161 has 79.11 GiB memory in use. Of the allocated memory 78.43 GiB is allocated by PyTorch, and 176.37 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":44856.80112822346,"meanTps":44796.10703158301,"stepMs":730.5023803710938,"jitter":0.000741699433791108,"achievedTflops":440.8469572343412,"nominalPeakTflops":2250.0,"mfuNominalPct":19.59319809930405,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.59319809930405,"vramAllocatedGb":81.15758592,"vramAllocatedPct":42.3792748093497,"vramReservedGb":83.52956416,"vramReservedPct":43.617886290152576,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":46446.97605281659,"meanTps":46390.0778508553,"stepMs":1410.9852905273438,"jitter":0.0004126958687679417,"achievedTflops":456.47499488627614,"nominalPeakTflops":2250.0,"mfuNominalPct":20.28777755050116,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.28777755050116,"vramAllocatedGb":155.164260352,"vramAllocatedPct":81.02445083234547,"vramReservedGb":160.293715968,"vramReservedPct":83.70297566411051,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.271528448,"vramAllocatedPct":99.35693997159238,"vramReservedGb":190.532550656,"vramReservedPct":99.49324185586823,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 191.81 MiB is free. Including non-PyTorch memory, this process has 178.15 GiB memory in use. Of the allocated memory 177.20 GiB is allocated by PyTorch, and 130.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":44900.363880755765,"meanTps":44886.87653874586,"stepMs":729.7936401367188,"jitter":0.0001928121402977802,"achievedTflops":441.2750864459567,"nominalPeakTflops":2250.0,"mfuNominalPct":19.612226064264743,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.612226064264743,"vramAllocatedGb":81.15758592,"vramAllocatedPct":42.3792748093497,"vramReservedGb":83.52956416,"vramReservedPct":43.617886290152576,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":46500.80354433447,"meanTps":46500.38829798024,"stepMs":1409.3519897460938,"jitter":0.00014080666592091584,"achievedTflops":457.00400465189404,"nominalPeakTflops":2250.0,"mfuNominalPct":20.311289095639737,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.311289095639737,"vramAllocatedGb":155.164260352,"vramAllocatedPct":81.02445083234547,"vramReservedGb":160.293715968,"vramReservedPct":83.70297566411051,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.271528448,"vramAllocatedPct":99.35693997159238,"vramReservedGb":190.532550656,"vramReservedPct":99.49324185586823,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 191.81 MiB is free. Including non-PyTorch memory, this process has 178.15 GiB memory in use. Of the allocated memory 177.20 GiB is allocated by PyTorch, and 130.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":44771.45946270504,"meanTps":44774.25128062422,"stepMs":731.8948364257812,"jitter":0.0005716447710498346,"achievedTflops":440.008230160122,"nominalPeakTflops":2250.0,"mfuNominalPct":19.555921340449867,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.555921340449867,"vramAllocatedGb":81.15758592,"vramAllocatedPct":28.23573381662647,"vramReservedGb":83.52956416,"vramReservedPct":29.06097455591472,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":46898.2894474835,"meanTps":46898.73661681674,"stepMs":1397.4070434570312,"jitter":0.00015042857391818525,"achievedTflops":460.91044573862905,"nominalPeakTflops":2250.0,"mfuNominalPct":20.484908699494625,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.484908699494625,"vramAllocatedGb":155.164260352,"vramAllocatedPct":53.983576562657824,"vramReservedGb":160.293715968,"vramReservedPct":55.76817798661274,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":278.01331968,"vramAllocatedPct":96.72429265822551,"vramReservedGb":279.680385024,"vramReservedPct":97.3042854312302,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.75 GiB is free. Including non-PyTorch memory, this process has 259.92 GiB memory in use. Of the allocated memory 258.92 GiB is allocated by PyTorch, and 189.84 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":44740.21390214346,"meanTps":44738.09513452646,"stepMs":732.4059753417969,"jitter":0.0005409136423336523,"achievedTflops":439.7011527503602,"nominalPeakTflops":2250.0,"mfuNominalPct":19.542273455571564,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.542273455571564,"vramAllocatedGb":81.15758592,"vramAllocatedPct":28.23573381662647,"vramReservedGb":83.52956416,"vramReservedPct":29.06097455591472,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":46872.448826136875,"meanTps":46854.27796233328,"stepMs":1398.1774291992188,"jitter":0.0005897977884242525,"achievedTflops":460.656487386558,"nominalPeakTflops":2250.0,"mfuNominalPct":20.4736216616248,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.4736216616248,"vramAllocatedGb":155.164260352,"vramAllocatedPct":53.983576562657824,"vramReservedGb":160.293715968,"vramReservedPct":55.76817798661274,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":278.01331968,"vramAllocatedPct":96.72429265822551,"vramReservedGb":279.680385024,"vramReservedPct":97.3042854312302,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.75 GiB is free. Including non-PyTorch memory, this process has 259.92 GiB memory in use. Of the allocated memory 258.92 GiB is allocated by PyTorch, and 189.84 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.963554816,"vramAllocatedPct":97.90517681646183,"vramReservedGb":33.00917248,"vramReservedPct":98.04066601005167,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 80.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 59.88 MiB is free. Including non-PyTorch memory, this process has 31.29 GiB memory in use. Of the allocated memory 30.66 GiB is allocated by PyTorch, and 43.50 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.963554816,"vramAllocatedPct":97.90517681646183,"vramReservedGb":33.00917248,"vramReservedPct":98.04066601005167,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 80.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 59.88 MiB is free. Including non-PyTorch memory, this process has 31.29 GiB memory in use. Of the allocated memory 30.66 GiB is allocated by PyTorch, and 43.50 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":35002.05349117296,"meanTps":35003.35086084123,"stepMs":468.0868225097656,"jitter":0.0005495131817025768,"achievedTflops":343.99574625102946,"nominalPeakTflops":989.5,"mfuNominalPct":34.76460295614245,"configuredPeakTflops":989.5,"mfuConfiguredPct":34.76460295614245,"vramAllocatedGb":44.204187136,"vramAllocatedPct":51.9942253224864,"vramReservedGb":44.335890432,"vramReservedPct":52.14913849455469,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":36217.062931854874,"meanTps":36188.88268788996,"stepMs":904.7669067382812,"jitter":0.0005986428288050331,"achievedTflops":355.936705068055,"nominalPeakTflops":989.5,"mfuNominalPct":35.97136989065741,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.97136989065741,"vramAllocatedGb":81.207655424,"vramAllocatedPct":95.5187598187416,"vramReservedGb":83.688947712,"vramReservedPct":98.43732655804833,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.931304448,"vramAllocatedPct":98.72239346135405,"vramReservedGb":84.129349632,"vramReservedPct":98.95533985372292,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 96.19 MiB is free. Including non-PyTorch memory, this process has 79.08 GiB memory in use. Of the allocated memory 78.17 GiB is allocated by PyTorch, and 188.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":34989.88891232609,"meanTps":34978.78233616777,"stepMs":468.2495574951172,"jitter":0.00036292949865779556,"achievedTflops":343.8761943116179,"nominalPeakTflops":989.5,"mfuNominalPct":34.75252090061828,"configuredPeakTflops":989.5,"mfuConfiguredPct":34.75252090061828,"vramAllocatedGb":44.204187136,"vramAllocatedPct":51.9942253224864,"vramReservedGb":44.335890432,"vramReservedPct":52.14913849455469,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":36221.58216873243,"meanTps":36205.267650798356,"stepMs":904.6540222167969,"jitter":0.00034579675825250304,"achievedTflops":355.98111955540986,"nominalPeakTflops":989.5,"mfuNominalPct":35.97585846947043,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.97585846947043,"vramAllocatedGb":81.207655424,"vramAllocatedPct":95.5187598187416,"vramReservedGb":83.688947712,"vramReservedPct":98.43732655804833,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.931304448,"vramAllocatedPct":98.72239346135405,"vramReservedGb":84.129349632,"vramReservedPct":98.95533985372292,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 96.19 MiB is free. Including non-PyTorch memory, this process has 79.08 GiB memory in use. Of the allocated memory 78.17 GiB is allocated by PyTorch, and 188.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":37173.00957573632,"meanTps":37126.787632995176,"stepMs":440.74989318847656,"jitter":0.0017714730957970031,"achievedTflops":365.33162754656314,"nominalPeakTflops":989.5,"mfuNominalPct":36.92083148525145,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.92083148525145,"vramAllocatedGb":44.204187136,"vramAllocatedPct":29.445752435104975,"vramReservedGb":44.335890432,"vramReservedPct":29.533484003089065,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":38161.49791674841,"meanTps":38164.64082353023,"stepMs":858.66650390625,"jitter":0.00034345029623854663,"achievedTflops":375.04636570078736,"nominalPeakTflops":989.5,"mfuNominalPct":37.902614017259964,"configuredPeakTflops":989.5,"mfuConfiguredPct":37.902614017259964,"vramAllocatedGb":81.207655424,"vramAllocatedPct":54.09488721268664,"vramReservedGb":83.647004672,"vramReservedPct":55.719811879627756,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.63218432,"vramAllocatedPct":95.01163262754724,"vramReservedGb":143.373893632,"vramReservedPct":95.50570774112799,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.12 GiB is free. Including non-PyTorch memory, this process has 133.68 GiB memory in use. Of the allocated memory 132.84 GiB is allocated by PyTorch, and 127.35 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":37122.04368856147,"meanTps":37108.80097480404,"stepMs":441.3550109863281,"jitter":0.0027840932862712698,"achievedTflops":364.83074126580595,"nominalPeakTflops":989.5,"mfuNominalPct":36.87021134571056,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.87021134571056,"vramAllocatedGb":44.204187136,"vramAllocatedPct":29.445752435104975,"vramReservedGb":44.335890432,"vramReservedPct":29.533484003089065,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":38172.02830585436,"meanTps":38164.34659912545,"stepMs":858.4296264648438,"jitter":0.0014109171445775824,"achievedTflops":375.1498570305097,"nominalPeakTflops":989.5,"mfuNominalPct":37.91307296922786,"configuredPeakTflops":989.5,"mfuConfiguredPct":37.91307296922786,"vramAllocatedGb":81.207655424,"vramAllocatedPct":54.09488721268664,"vramReservedGb":83.647004672,"vramReservedPct":55.719811879627756,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.63218432,"vramAllocatedPct":95.01163262754724,"vramReservedGb":143.373893632,"vramReservedPct":95.50570774112799,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.12 GiB is free. Including non-PyTorch memory, this process has 133.68 GiB memory in use. Of the allocated memory 132.84 GiB is allocated by PyTorch, and 127.35 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":9883.496042153212,"meanTps":9886.543583499371,"stepMs":1657.7130126953125,"jitter":0.0011140255604111492,"achievedTflops":97.13374666566331,"nominalPeakTflops":154.8,"mfuNominalPct":62.74789836283159,"configuredPeakTflops":180.6,"mfuConfiguredPct":53.78391288242708,"vramAllocatedGb":44.154117632,"vramAllocatedPct":86.51913755303602,"vramReservedGb":44.293947392,"vramReservedPct":86.79313125709501,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.583312896,"vramAllocatedPct":99.11702104008393,"vramReservedGb":50.679775232,"vramReservedPct":99.3060371175114,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 16.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 2.25 MiB is free. Process 1472317 has 47.52 GiB memory in use. Of the allocated memory 47.11 GiB is allocated by PyTorch, and 91.99 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":9895.741489247304,"meanTps":9894.59125740931,"stepMs":1655.6616821289062,"jitter":0.001088023498618189,"achievedTflops":97.25409336796099,"nominalPeakTflops":154.8,"mfuNominalPct":62.825641710569116,"configuredPeakTflops":180.6,"mfuConfiguredPct":53.850550037630676,"vramAllocatedGb":44.154117632,"vramAllocatedPct":86.51913755303602,"vramReservedGb":44.293947392,"vramReservedPct":86.79313125709501,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.583312896,"vramAllocatedPct":99.11702104008393,"vramReservedGb":50.679775232,"vramReservedPct":99.3060371175114,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 16.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 2.25 MiB is free. Process 1479887 has 47.52 GiB memory in use. Of the allocated memory 47.11 GiB is allocated by PyTorch, and 91.99 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":28361.12530829009,"meanTps":28358.69779227646,"stepMs":577.6921691894531,"jitter":0.0001944707547885962,"achievedTflops":278.72954560807597,"nominalPeakTflops":468.0,"mfuNominalPct":59.557595215400845,"configuredPeakTflops":468.0,"mfuConfiguredPct":59.557595215400845,"vramAllocatedGb":44.201558016,"vramAllocatedPct":43.34593081358022,"vramReservedGb":44.539314176,"vramReservedPct":43.67714889276921,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":28231.166730241504,"meanTps":28231.72744681124,"stepMs":1160.7030029296875,"jitter":0.00010493684636628003,"achievedTflops":277.4523291713655,"nominalPeakTflops":468.0,"mfuNominalPct":59.28468572037724,"configuredPeakTflops":468.0,"mfuConfiguredPct":59.28468572037724,"vramAllocatedGb":81.202921472,"vramAllocatedPct":79.63104410735482,"vramReservedGb":83.751862272,"vramReservedPct":82.13064404283037,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":28365.625191857438,"meanTps":28364.940302276013,"stepMs":577.6005249023438,"jitter":0.00020061160848307592,"achievedTflops":278.7737698935506,"nominalPeakTflops":468.0,"mfuNominalPct":59.567044849049275,"configuredPeakTflops":468.0,"mfuConfiguredPct":59.567044849049275,"vramAllocatedGb":44.201558016,"vramAllocatedPct":43.34593081358022,"vramReservedGb":44.539314176,"vramReservedPct":43.67714889276921,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":28225.30852238629,"meanTps":28225.439107058337,"stepMs":1160.9439086914062,"jitter":0.0003201234746716836,"achievedTflops":277.3947554469165,"nominalPeakTflops":468.0,"mfuNominalPct":59.272383642503534,"configuredPeakTflops":468.0,"mfuConfiguredPct":59.272383642503534,"vramAllocatedGb":81.202921472,"vramAllocatedPct":79.63104410735482,"vramReservedGb":83.751862272,"vramReservedPct":82.13064404283037,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":11990.395907294678,"meanTps":11986.232602728594,"stepMs":2732.8538818359375,"jitter":0.001313162442732093,"achievedTflops":199.31210368911326,"nominalPeakTflops":312.0,"mfuNominalPct":63.882084515741425,"configuredPeakTflops":312.0,"mfuConfiguredPct":63.882084515741425,"vramAllocatedGb":81.21426688,"vramAllocatedPct":95.57516151326851,"vramReservedGb":81.730207744,"vramReservedPct":96.18233477607659,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.105630208,"vramAllocatedPct":98.97779663740819,"vramReservedGb":84.297121792,"vramReservedPct":99.2031491496247,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 280.75 MiB is free. Process 949988 has 78.86 GiB memory in use. Of the allocated memory 78.17 GiB is allocated by PyTorch, and 182.62 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":11965.433557734328,"meanTps":11962.977158745274,"stepMs":2738.55517578125,"jitter":0.0012824740428070587,"achievedTflops":198.89716339503426,"nominalPeakTflops":312.0,"mfuNominalPct":63.749090831741746,"configuredPeakTflops":312.0,"mfuConfiguredPct":63.749090831741746,"vramAllocatedGb":81.21426688,"vramAllocatedPct":95.57516151326851,"vramReservedGb":81.730207744,"vramReservedPct":96.18233477607659,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.105630208,"vramAllocatedPct":98.97779663740819,"vramReservedGb":84.297121792,"vramReservedPct":99.2031491496247,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 280.75 MiB is free. Process 970883 has 78.86 GiB memory in use. Of the allocated memory 78.17 GiB is allocated by PyTorch, and 182.62 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":28555.93508680565,"meanTps":28527.6530616485,"stepMs":1147.5022583007812,"jitter":0.00614460369698389,"achievedTflops":474.6751932935252,"nominalPeakTflops":2250.0,"mfuNominalPct":21.096675257490006,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.096675257490006,"vramAllocatedGb":81.21426688,"vramAllocatedPct":42.408872757010016,"vramReservedGb":81.688264704,"vramReservedPct":42.6563873154412,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":29310.078064899382,"meanTps":29318.03319084929,"stepMs":2235.9544677734375,"jitter":0.0032069820508765137,"achievedTflops":487.2110448707688,"nominalPeakTflops":2250.0,"mfuNominalPct":21.65382421647861,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.65382421647861,"vramAllocatedGb":155.221523456,"vramAllocatedPct":81.05435276687619,"vramReservedGb":160.3796992,"vramReservedPct":83.74787481896605,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.328151552,"vramAllocatedPct":99.38650770771677,"vramReservedGb":190.578688,"vramReservedPct":99.51733408530292,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 89.81 MiB is free. Including non-PyTorch memory, this process has 178.25 GiB memory in use. Of the allocated memory 177.26 GiB is allocated by PyTorch, and 178.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":28566.507352802284,"meanTps":28558.98471197116,"stepMs":1147.0775756835938,"jitter":0.003618043017457056,"achievedTflops":474.85093232606766,"nominalPeakTflops":2250.0,"mfuNominalPct":21.10448588115856,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.10448588115856,"vramAllocatedGb":81.21426688,"vramAllocatedPct":42.408872757010016,"vramReservedGb":81.688264704,"vramReservedPct":42.6563873154412,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":29375.246842865323,"meanTps":29361.980511579808,"stepMs":2230.9940185546875,"jitter":0.000840174744312348,"achievedTflops":488.2943223815086,"nominalPeakTflops":2250.0,"mfuNominalPct":21.701969883622606,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.701969883622606,"vramAllocatedGb":155.221523456,"vramAllocatedPct":81.05435276687619,"vramReservedGb":160.3796992,"vramReservedPct":83.74787481896605,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.328151552,"vramAllocatedPct":99.38650770771677,"vramReservedGb":190.578688,"vramReservedPct":99.51733408530292,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 89.75 MiB is free. Including non-PyTorch memory, this process has 178.25 GiB memory in use. Of the allocated memory 177.26 GiB is allocated by PyTorch, and 178.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":28820.01502544099,"meanTps":28812.830071091474,"stepMs":1136.9876098632812,"jitter":0.0004112318902235591,"achievedTflops":479.0649005657829,"nominalPeakTflops":2250.0,"mfuNominalPct":21.291773358479244,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.291773358479244,"vramAllocatedGb":81.21426688,"vramAllocatedPct":28.255453827773778,"vramReservedGb":81.688264704,"vramReservedPct":28.42036356771253,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":29712.207209737542,"meanTps":29708.955985221946,"stepMs":2205.6927490234375,"jitter":0.0003388746548999435,"achievedTflops":493.89549519518516,"nominalPeakTflops":2250.0,"mfuNominalPct":21.950910897563784,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.950910897563784,"vramAllocatedGb":155.221523456,"vramAllocatedPct":54.003499108945135,"vramReservedGb":160.3796992,"vramReservedPct":55.79809262273608,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":278.070454784,"vramAllocatedPct":96.74417067172038,"vramReservedGb":279.724425216,"vramReservedPct":97.31960756192751,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.65 GiB is free. Including non-PyTorch memory, this process has 260.02 GiB memory in use. Of the allocated memory 258.97 GiB is allocated by PyTorch, and 237.35 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":28831.80256217929,"meanTps":28811.6702162378,"stepMs":1136.5227661132812,"jitter":0.0002760449194714574,"achievedTflops":479.2608406133667,"nominalPeakTflops":2250.0,"mfuNominalPct":21.300481805038523,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.300481805038523,"vramAllocatedGb":81.21426688,"vramAllocatedPct":28.255453827773778,"vramReservedGb":81.688264704,"vramReservedPct":28.42036356771253,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":29723.467198606595,"meanTps":29722.002670640886,"stepMs":2204.857177734375,"jitter":8.43813956114956e-05,"achievedTflops":494.082666001417,"nominalPeakTflops":2250.0,"mfuNominalPct":21.95922960006298,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.95922960006298,"vramAllocatedGb":155.221523456,"vramAllocatedPct":54.003499108945135,"vramReservedGb":160.3796992,"vramReservedPct":55.79809262273608,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":278.070454784,"vramAllocatedPct":96.74417067172038,"vramReservedGb":279.724425216,"vramReservedPct":97.31960756192751,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 15.62 GiB. GPU 0 has a total capacity of 267.69 GiB of which 7.65 GiB is free. Including non-PyTorch memory, this process has 260.02 GiB memory in use. Of the allocated memory 258.97 GiB is allocated by PyTorch, and 237.35 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.967210496,"vramAllocatedPct":97.91603456523262,"vramReservedGb":33.011269632,"vramReservedPct":98.04689476901036,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 80.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 77.88 MiB is free. Including non-PyTorch memory, this process has 31.27 GiB memory in use. Of the allocated memory 30.61 GiB is allocated by PyTorch, and 74.27 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.967210496,"vramAllocatedPct":97.91603456523262,"vramReservedGb":33.011269632,"vramReservedPct":98.04689476901036,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 80.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 77.88 MiB is free. Including non-PyTorch memory, this process has 31.27 GiB memory in use. Of the allocated memory 30.61 GiB is allocated by PyTorch, and 74.27 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_32k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":24425.57044918187,"meanTps":24419.676362672213,"stepMs":1341.544921875,"jitter":0.0007355033914665199,"achievedTflops":406.0176050626743,"nominalPeakTflops":989.5,"mfuNominalPct":41.03260283604592,"configuredPeakTflops":989.5,"mfuConfiguredPct":41.03260283604592,"vramAllocatedGb":81.264336384,"vramAllocatedPct":95.58542958006235,"vramReservedGb":81.751179264,"vramReservedPct":96.15806805708013,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_32k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.987927552,"vramAllocatedPct":98.78899517079793,"vramReservedGb":84.129349632,"vramReservedPct":98.95533985372292,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 96.19 MiB is free. Including non-PyTorch memory, this process has 79.08 GiB memory in use. Of the allocated memory 78.22 GiB is allocated by PyTorch, and 134.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_32k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":24418.87244011514,"meanTps":24416.04236085879,"stepMs":1341.9129028320312,"jitter":0.0008699002146875854,"achievedTflops":405.9062664306608,"nominalPeakTflops":989.5,"mfuNominalPct":41.021350826746925,"configuredPeakTflops":989.5,"mfuConfiguredPct":41.021350826746925,"vramAllocatedGb":81.264336384,"vramAllocatedPct":95.58542958006235,"vramReservedGb":81.751179264,"vramReservedPct":96.15806805708013,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_32k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.987927552,"vramAllocatedPct":98.78899517079793,"vramReservedGb":84.129349632,"vramReservedPct":98.95533985372292,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 96.19 MiB is free. Including non-PyTorch memory, this process has 79.08 GiB memory in use. Of the allocated memory 78.22 GiB is allocated by PyTorch, and 134.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":25345.21984215715,"meanTps":25346.04738676922,"stepMs":1292.8670654296875,"jitter":0.001404465230318812,"achievedTflops":421.30461114550144,"nominalPeakTflops":989.5,"mfuNominalPct":42.57752512839832,"configuredPeakTflops":989.5,"mfuConfiguredPct":42.57752512839832,"vramAllocatedGb":81.264336384,"vramAllocatedPct":54.132644122700825,"vramReservedGb":81.751179264,"vramReservedPct":54.45694496042845,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.689063424,"vramAllocatedPct":95.04952152730151,"vramReservedGb":143.53956864,"vramReservedPct":95.61606889888988,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.00 GiB is free. Including non-PyTorch memory, this process has 133.80 GiB memory in use. Of the allocated memory 132.89 GiB is allocated by PyTorch, and 191.10 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":25298.42053247932,"meanTps":25298.539505783854,"stepMs":1295.2587280273438,"jitter":0.0016458336841957447,"achievedTflops":420.5266827989143,"nominalPeakTflops":989.5,"mfuNominalPct":42.49890680130513,"configuredPeakTflops":989.5,"mfuConfiguredPct":42.49890680130513,"vramAllocatedGb":81.264336384,"vramAllocatedPct":54.132644122700825,"vramReservedGb":81.751179264,"vramReservedPct":54.45694496042845,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":142.689063424,"vramAllocatedPct":95.04952152730151,"vramReservedGb":143.53956864,"vramReservedPct":95.61606889888988,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 7.81 GiB. GPU 0 has a total capacity of 139.81 GiB of which 6.00 GiB is free. Including non-PyTorch memory, this process has 133.80 GiB memory in use. Of the allocated memory 132.89 GiB is allocated by PyTorch, and 191.10 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.549478912,"vramAllocatedPct":99.0507240043225,"vramReservedGb":50.639929344,"vramReservedPct":99.22795987240535,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 140.25 MiB is free. Process 1490581 has 47.38 GiB memory in use. Of the allocated memory 47.00 GiB is allocated by PyTorch, and 66.26 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.549478912,"vramAllocatedPct":99.0507240043225,"vramReservedGb":50.639929344,"vramReservedPct":99.22795987240535,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 140.25 MiB is free. Process 1493539 has 47.38 GiB memory in use. Of the allocated memory 47.00 GiB is allocated by PyTorch, and 66.26 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":21516.865804312878,"meanTps":21516.09272716062,"stepMs":1522.8983764648438,"jitter":0.0002364324442219476,"achievedTflops":357.6672381305505,"nominalPeakTflops":468.0,"mfuNominalPct":76.42462353216892,"configuredPeakTflops":468.0,"mfuConfiguredPct":76.42462353216892,"vramAllocatedGb":81.262560768,"vramAllocatedPct":79.68952894169598,"vramReservedGb":86.027272192,"vramReservedPct":84.36200794473518,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.3740288,"vramAllocatedPct":98.43104865828963,"vramReservedGb":100.94641152,"vramReservedPct":98.992351537501,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 711.75 MiB is free. Including non-PyTorch memory, this process has 94.27 GiB memory in use. Of the allocated memory 93.48 GiB is allocated by PyTorch, and 145.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":21515.843957315126,"meanTps":21515.270213373075,"stepMs":1522.970703125,"jitter":0.00018342103381677205,"achievedTflops":357.6502523298859,"nominalPeakTflops":468.0,"mfuNominalPct":76.42099408758246,"configuredPeakTflops":468.0,"mfuConfiguredPct":76.42099408758246,"vramAllocatedGb":81.262560768,"vramAllocatedPct":79.68952894169598,"vramReservedGb":86.027272192,"vramReservedPct":84.36200794473518,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"500m","modelLabel":"500M","parameters":505516800,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.3740288,"vramAllocatedPct":98.43104865828963,"vramReservedGb":100.94641152,"vramReservedPct":98.992351537501,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 711.75 MiB is free. Including non-PyTorch memory, this process has 94.27 GiB memory in use. Of the allocated memory 93.48 GiB is allocated by PyTorch, and 145.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":21848.538247899727,"meanTps":21833.18560329987,"stepMs":374.9449920654297,"jitter":0.002929976447147297,"achievedTflops":98.3115256681479,"nominalPeakTflops":312.0,"mfuNominalPct":31.510104380816635,"configuredPeakTflops":312.0,"mfuConfiguredPct":31.510104380816635,"vramAllocatedGb":31.389135872,"vramAllocatedPct":36.93959012350725,"vramReservedGb":32.004636672,"vramReservedPct":37.66392823097877,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23230.809712224815,"meanTps":23230.137042041973,"stepMs":705.2702941894531,"jitter":0.0013160391837456697,"achievedTflops":104.53131094638762,"nominalPeakTflops":312.0,"mfuNominalPct":33.503625303329365,"configuredPeakTflops":312.0,"mfuConfiguredPct":33.503625303329365,"vramAllocatedGb":52.854756352,"vramAllocatedPct":62.20091702053981,"vramReservedGb":54.133784576,"vramReservedPct":63.70611227483488,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.424628224,"vramAllocatedPct":98.1763749523756,"vramReservedGb":83.512786944,"vramReservedPct":98.280122535485,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.14 GiB of which 868.75 MiB is free. Process 764857 has 78.28 GiB memory in use. Of the allocated memory 77.70 GiB is allocated by PyTorch, and 84.07 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":21829.857193933516,"meanTps":21828.142250208286,"stepMs":375.26585388183594,"jitter":0.0022621724922146868,"achievedTflops":98.22746682193728,"nominalPeakTflops":312.0,"mfuNominalPct":31.483162442928617,"configuredPeakTflops":312.0,"mfuConfiguredPct":31.483162442928617,"vramAllocatedGb":31.389135872,"vramAllocatedPct":36.93959012350725,"vramReservedGb":32.004636672,"vramReservedPct":37.66392823097877,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23227.44775574529,"meanTps":23236.334881695457,"stepMs":705.3723754882812,"jitter":0.0018946836158490296,"achievedTflops":104.51618320342457,"nominalPeakTflops":312.0,"mfuNominalPct":33.49877666776428,"configuredPeakTflops":312.0,"mfuConfiguredPct":33.49877666776428,"vramAllocatedGb":52.854756352,"vramAllocatedPct":62.20091702053981,"vramReservedGb":54.133784576,"vramReservedPct":63.70611227483488,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.424628224,"vramAllocatedPct":98.1763749523756,"vramReservedGb":83.512786944,"vramReservedPct":98.280122535485,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.14 GiB of which 868.75 MiB is free. Process 852916 has 78.28 GiB memory in use. Of the allocated memory 77.70 GiB is allocated by PyTorch, and 84.07 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":84813.22102647726,"meanTps":84813.48497839256,"stepMs":386.3548583984375,"jitter":0.0009128006577527654,"achievedTflops":381.63272349555706,"nominalPeakTflops":2250.0,"mfuNominalPct":16.961454377580313,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.961454377580313,"vramAllocatedGb":95.7859968,"vramAllocatedPct":50.01801168995012,"vramReservedGb":98.431926272,"vramReservedPct":51.399676397554636,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":89150.26377885418,"meanTps":89077.06547623572,"stepMs":735.118408203125,"jitter":0.001074942793946952,"achievedTflops":401.1480469023826,"nominalPeakTflops":2250.0,"mfuNominalPct":17.828802084550336,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.828802084550336,"vramAllocatedGb":181.648478208,"vramAllocatedPct":94.85411239641026,"vramReservedGb":186.816397312,"vramReservedPct":97.55272228503846,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.174687232,"vramAllocatedPct":98.78418593032623,"vramReservedGb":190.266212352,"vramReservedPct":99.35416398594985,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.14 GiB is free. Including non-PyTorch memory, this process has 177.20 GiB memory in use. Of the allocated memory 176.18 GiB is allocated by PyTorch, and 200.96 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":84796.6569764113,"meanTps":84775.15145319494,"stepMs":386.4303283691406,"jitter":0.0005042284061447659,"achievedTflops":381.55819049866955,"nominalPeakTflops":2250.0,"mfuNominalPct":16.95814179994087,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.95814179994087,"vramAllocatedGb":95.7859968,"vramAllocatedPct":50.01801168995012,"vramReservedGb":98.431926272,"vramReservedPct":51.399676397554636,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":89054.61505056433,"meanTps":89025.22969762,"stepMs":735.907958984375,"jitter":0.0009605389796365897,"achievedTflops":400.717657816408,"nominalPeakTflops":2250.0,"mfuNominalPct":17.809673680729247,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.809673680729247,"vramAllocatedGb":181.648478208,"vramAllocatedPct":94.85411239641026,"vramReservedGb":186.816397312,"vramReservedPct":97.55272228503846,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.174687232,"vramAllocatedPct":98.78418593032623,"vramReservedGb":190.266212352,"vramReservedPct":99.35416398594985,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.14 GiB is free. Including non-PyTorch memory, this process has 177.20 GiB memory in use. Of the allocated memory 176.18 GiB is allocated by PyTorch, and 200.96 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":82136.12179845016,"meanTps":82034.9764582273,"stepMs":398.94749450683594,"jitter":0.0004826239988544075,"achievedTflops":369.5866219904523,"nominalPeakTflops":2250.0,"mfuNominalPct":16.426072088464544,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.426072088464544,"vramAllocatedGb":95.7859968,"vramAllocatedPct":33.3251399526724,"vramReservedGb":98.431926272,"vramReservedPct":34.24569173377889,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":89288.86183338393,"meanTps":89288.60010395809,"stepMs":733.9773254394531,"jitter":0.0003060628866969975,"achievedTflops":401.7716943995677,"nominalPeakTflops":2250.0,"mfuNominalPct":17.8565197510919,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.8565197510919,"vramAllocatedGb":181.648478208,"vramAllocatedPct":63.19776544280388,"vramReservedGb":186.816397312,"vramReservedPct":64.99574879275518,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.27230976,"vramAllocatedPct":99.59769805649489,"vramReservedGb":286.506614784,"vramReservedPct":99.67921568931462,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 384.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 25.62 MiB is free. Including non-PyTorch memory, this process has 267.65 GiB memory in use. Of the allocated memory 266.61 GiB is allocated by PyTorch, and 223.45 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":82296.34289895569,"meanTps":82286.64806413544,"stepMs":398.17079162597656,"jitter":0.0002694807772098978,"achievedTflops":370.3075663692572,"nominalPeakTflops":2250.0,"mfuNominalPct":16.458114060855873,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.458114060855873,"vramAllocatedGb":95.7859968,"vramAllocatedPct":33.3251399526724,"vramReservedGb":98.431926272,"vramReservedPct":34.24569173377889,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":89382.74896063408,"meanTps":89382.60275322238,"stepMs":733.2063598632812,"jitter":0.00017957291244636057,"achievedTflops":402.1941568369094,"nominalPeakTflops":2250.0,"mfuNominalPct":17.8752958594182,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.8752958594182,"vramAllocatedGb":181.648478208,"vramAllocatedPct":63.19776544280388,"vramReservedGb":186.816397312,"vramReservedPct":64.99574879275518,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.27230976,"vramAllocatedPct":99.59769805649489,"vramReservedGb":286.506614784,"vramReservedPct":99.67921568931462,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 384.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 25.62 MiB is free. Including non-PyTorch memory, this process has 267.65 GiB memory in use. Of the allocated memory 266.61 GiB is allocated by PyTorch, and 223.45 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15381.20266492843,"meanTps":15382.360296619378,"stepMs":133.1495361328125,"jitter":0.0009458573188810381,"achievedTflops":69.21055694631916,"nominalPeakTflops":165.2,"mfuNominalPct":41.89501025806245,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":33.676371882078904,"vramAllocatedGb":15.289920512,"vramAllocatedPct":60.52765755487699,"vramReservedGb":15.548284928,"vramReservedPct":61.55043546165162,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":17205.845881878624,"meanTps":17206.236075221626,"stepMs":238.05862426757812,"jitter":0.0004512018336804173,"achievedTflops":77.4208754776129,"nominalPeakTflops":165.2,"mfuNominalPct":46.864936729789896,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":37.6713367014579,"vramAllocatedGb":20.656325632,"vramAllocatedPct":81.7714521939121,"vramReservedGb":21.084766208,"vramReservedPct":83.46750446876791,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.647318016,"vramAllocatedPct":97.57044998028294,"vramReservedGb":24.733810688,"vramReservedPct":97.91284540527639,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 24.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 23.69 MiB is free. Including non-PyTorch memory, this process has 23.49 GiB memory in use. Of the allocated memory 22.95 GiB is allocated by PyTorch, and 82.49 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15362.853166604675,"meanTps":15356.705745731711,"stepMs":133.3085708618164,"jitter":0.0015943467360337975,"achievedTflops":69.1279899958447,"nominalPeakTflops":165.2,"mfuNominalPct":41.84503026382851,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":33.63619657570904,"vramAllocatedGb":15.289920512,"vramAllocatedPct":60.52765755487699,"vramReservedGb":15.548284928,"vramReservedPct":61.55043546165162,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":17179.446432182198,"meanTps":17174.393822375245,"stepMs":238.42444610595703,"jitter":0.0006084915094409357,"achievedTflops":77.30208628691246,"nominalPeakTflops":165.2,"mfuNominalPct":46.79303044001965,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":37.61353643025503,"vramAllocatedGb":20.656325632,"vramAllocatedPct":81.7714521939121,"vramReservedGb":21.084766208,"vramReservedPct":83.46750446876791,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.647318016,"vramAllocatedPct":97.57044998028294,"vramReservedGb":24.733810688,"vramReservedPct":97.91284540527639,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 24.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 23.69 MiB is free. Including non-PyTorch memory, this process has 23.49 GiB memory in use. Of the allocated memory 22.95 GiB is allocated by PyTorch, and 82.49 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":23120.020119684563,"meanTps":23108.67856577438,"stepMs":177.1624755859375,"jitter":0.000466212762219091,"achievedTflops":104.03279275046978,"nominalPeakTflops":209.5,"mfuNominalPct":49.65765763745574,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":38.68154755124789,"vramAllocatedGb":20.656325632,"vramAllocatedPct":61.35142962183647,"vramReservedGb":21.019754496,"vramReservedPct":62.43085104312248,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":26577.02850049419,"meanTps":26578.744887033594,"stepMs":308.2361145019531,"jitter":0.00021291598718887125,"achievedTflops":119.58823926633174,"nominalPeakTflops":209.5,"mfuNominalPct":57.08269177390537,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":44.465384821938585,"vramAllocatedGb":31.389135872,"vramAllocatedPct":93.22898925247107,"vramReservedGb":32.052871168,"vramReservedPct":95.20035192488116,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.926181888,"vramAllocatedPct":97.79417534394429,"vramReservedGb":32.996589568,"vramReservedPct":98.00329345629942,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 48.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 31.88 MiB is free. Including non-PyTorch memory, this process has 31.32 GiB memory in use. Of the allocated memory 30.66 GiB is allocated by PyTorch, and 67.15 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":23119.2136716067,"meanTps":23115.498066707278,"stepMs":177.1686553955078,"jitter":0.00023520513654166228,"achievedTflops":104.02916398867312,"nominalPeakTflops":209.5,"mfuNominalPct":49.65592553158621,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":38.68019830243625,"vramAllocatedGb":20.656325632,"vramAllocatedPct":61.35142962183647,"vramReservedGb":21.019754496,"vramReservedPct":62.43085104312248,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":26576.881147629036,"meanTps":26577.4767526928,"stepMs":308.2378234863281,"jitter":0.0001882545602191293,"achievedTflops":119.58757622495017,"nominalPeakTflops":209.5,"mfuNominalPct":57.0823752863724,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":44.46513828941694,"vramAllocatedGb":31.389135872,"vramAllocatedPct":93.22898925247107,"vramReservedGb":32.052871168,"vramReservedPct":95.20035192488116,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.926181888,"vramAllocatedPct":97.79417534394429,"vramReservedGb":32.996589568,"vramReservedPct":98.00329345629942,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 48.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 31.88 MiB is free. Including non-PyTorch memory, this process has 31.32 GiB memory in use. Of the allocated memory 30.66 GiB is allocated by PyTorch, and 67.15 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":48224.87862171596,"meanTps":48220.170804398236,"stepMs":169.8708267211914,"jitter":0.00054502189931431,"achievedTflops":216.99673171123462,"nominalPeakTflops":989.5,"mfuNominalPct":21.929937515031288,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.929937515031288,"vramAllocatedGb":31.439205376,"vramAllocatedPct":36.979689802923694,"vramReservedGb":32.067551232,"vramReservedPct":37.718768115048285,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":52847.90348617203,"meanTps":52845.07000884516,"stepMs":310.0217590332031,"jitter":0.0003776263085585069,"achievedTflops":237.79888435273452,"nominalPeakTflops":989.5,"mfuNominalPct":24.032226816850383,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.032226816850383,"vramAllocatedGb":52.904825856,"vramAllocatedPct":62.22816467002861,"vramReservedGb":54.196699136,"vramReservedPct":63.747702857706685,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.474697728,"vramAllocatedPct":98.18531961786498,"vramReservedGb":83.554729984,"vramReservedPct":98.27945583936655,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.18 GiB of which 644.19 MiB is free. Including non-PyTorch memory, this process has 78.54 GiB memory in use. Of the allocated memory 77.74 GiB is allocated by PyTorch, and 76.32 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":48218.67189385749,"meanTps":48198.26787550235,"stepMs":169.89269256591797,"jitter":0.0004360945609597553,"achievedTflops":216.96880339501274,"nominalPeakTflops":989.5,"mfuNominalPct":21.927115047500024,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.927115047500024,"vramAllocatedGb":31.439205376,"vramAllocatedPct":36.979689802923694,"vramReservedGb":32.067551232,"vramReservedPct":37.718768115048285,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":52824.322111812195,"meanTps":52811.25326836518,"stepMs":310.16015625,"jitter":0.0003838764133165829,"achievedTflops":237.69277561152893,"nominalPeakTflops":989.5,"mfuNominalPct":24.021503346288927,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.021503346288927,"vramAllocatedGb":52.904825856,"vramAllocatedPct":62.22816467002861,"vramReservedGb":54.196699136,"vramReservedPct":63.747702857706685,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.474697728,"vramAllocatedPct":98.18531961786498,"vramReservedGb":83.554729984,"vramReservedPct":98.27945583936655,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.18 GiB of which 644.19 MiB is free. Including non-PyTorch memory, this process has 78.54 GiB memory in use. Of the allocated memory 77.74 GiB is allocated by PyTorch, and 76.32 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":58319.607133323036,"meanTps":58245.43168718167,"stepMs":280.93467712402344,"jitter":0.003974223368376057,"achievedTflops":262.41982363260126,"nominalPeakTflops":989.5,"mfuNominalPct":26.520447057362432,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.520447057362432,"vramAllocatedGb":52.904825856,"vramAllocatedPct":35.24151230256245,"vramReservedGb":54.144270336,"vramReservedPct":36.06714393792883,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":61225.289684068826,"meanTps":61239.69873409121,"stepMs":535.2036743164062,"jitter":0.001947261767519005,"achievedTflops":275.49447793807155,"nominalPeakTflops":989.5,"mfuNominalPct":27.841786552609555,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.841786552609555,"vramAllocatedGb":95.836066304,"vramAllocatedPct":63.83931626340606,"vramReservedGb":98.341748736,"vramReservedPct":65.50842748010291,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.269484544,"vramAllocatedPct":98.76680962667544,"vramReservedGb":148.985872384,"vramReservedPct":99.24401733822683,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 773.44 MiB is free. Including non-PyTorch memory, this process has 139.05 GiB memory in use. Of the allocated memory 138.09 GiB is allocated by PyTorch, and 243.20 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":57927.10629417667,"meanTps":57956.03008356208,"stepMs":282.8382263183594,"jitter":0.006227754942440383,"achievedTflops":260.6536937485475,"nominalPeakTflops":989.5,"mfuNominalPct":26.341959954375692,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.341959954375692,"vramAllocatedGb":52.904825856,"vramAllocatedPct":35.24151230256245,"vramReservedGb":54.144270336,"vramReservedPct":36.06714393792883,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":61140.03032682478,"meanTps":61129.79991499092,"stepMs":535.9500122070312,"jitter":0.0019297764549729786,"achievedTflops":275.1108377424189,"nominalPeakTflops":989.5,"mfuNominalPct":27.80301543632328,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.80301543632328,"vramAllocatedGb":95.836066304,"vramAllocatedPct":63.83931626340606,"vramReservedGb":98.341748736,"vramReservedPct":65.50842748010291,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.269484544,"vramAllocatedPct":98.76680962667544,"vramReservedGb":148.985872384,"vramReservedPct":99.24401733822683,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 773.44 MiB is free. Including non-PyTorch memory, this process has 139.05 GiB memory in use. Of the allocated memory 138.09 GiB is allocated by PyTorch, and 243.20 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":15019.909998755034,"meanTps":15014.96904877761,"stepMs":272.70469665527344,"jitter":0.0011342669067072888,"achievedTflops":67.58485398984638,"nominalPeakTflops":364.2,"mfuNominalPct":18.55707138655859,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.97116387224775,"vramAllocatedGb":20.656325632,"vramAllocatedPct":40.60064427445419,"vramReservedGb":21.019754496,"vramReservedPct":41.31497490078176,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":15573.674637199874,"meanTps":15571.058907065906,"stepMs":526.015869140625,"jitter":0.003832837408641482,"achievedTflops":70.07662006814755,"nominalPeakTflops":364.2,"mfuNominalPct":19.241246586531453,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.523131304109915,"vramAllocatedGb":31.389135872,"vramAllocatedPct":61.69631338728023,"vramReservedGb":32.052871168,"vramReservedPct":63.00090555557703,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":49.715170816,"vramAllocatedPct":97.71669953814089,"vramReservedGb":49.952063488,"vramReservedPct":98.18231938159443,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.38 GiB of which 436.50 MiB is free. Process 2319972 has 46.95 GiB memory in use. Of the allocated memory 46.30 GiB is allocated by PyTorch, and 145.92 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":10053.793650369775,"meanTps":10054.889410769274,"stepMs":407.4084014892578,"jitter":0.0009873480355762903,"achievedTflops":45.23889796680583,"nominalPeakTflops":154.8,"mfuNominalPct":29.22409429380221,"configuredPeakTflops":180.6,"mfuConfiguredPct":25.049223680401898,"vramAllocatedGb":20.656325632,"vramAllocatedPct":40.475669644003716,"vramReservedGb":21.019754496,"vramReservedPct":41.18780145778435,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":11480.829013372331,"meanTps":11481.402956067339,"stepMs":713.5373229980469,"jitter":0.0006670726054972991,"achievedTflops":51.66010665945907,"nominalPeakTflops":154.8,"mfuNominalPct":33.37216192471516,"configuredPeakTflops":180.6,"mfuConfiguredPct":28.604710221184426,"vramAllocatedGb":31.389135872,"vramAllocatedPct":61.50640324906641,"vramReservedGb":32.052871168,"vramReservedPct":62.80697969477961,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.715170816,"vramAllocatedPct":97.41591346363758,"vramReservedGb":49.952063488,"vramReservedPct":97.8801000621536,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 754.25 MiB is free. Process 1415479 has 46.79 GiB memory in use. Of the allocated memory 46.30 GiB is allocated by PyTorch, and 165.92 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":10128.604820307419,"meanTps":10128.835619508414,"stepMs":404.39923095703125,"jitter":0.0010069461581058057,"achievedTflops":45.57552461752907,"nominalPeakTflops":154.8,"mfuNominalPct":29.441553370496816,"configuredPeakTflops":180.6,"mfuConfiguredPct":25.235617174711557,"vramAllocatedGb":20.656325632,"vramAllocatedPct":40.475669644003716,"vramReservedGb":21.019754496,"vramReservedPct":41.18780145778435,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":11503.289296210385,"meanTps":11507.645481731854,"stepMs":712.1441345214844,"jitter":0.0014913969803740123,"achievedTflops":51.76117084268696,"nominalPeakTflops":154.8,"mfuNominalPct":33.437448864784855,"configuredPeakTflops":180.6,"mfuConfiguredPct":28.66067045552988,"vramAllocatedGb":31.389135872,"vramAllocatedPct":61.50640324906641,"vramReservedGb":32.052871168,"vramReservedPct":62.80697969477961,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.715170816,"vramAllocatedPct":97.41591346363758,"vramReservedGb":49.952063488,"vramReservedPct":97.8801000621536,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 754.25 MiB is free. Process 1449495 has 46.79 GiB memory in use. Of the allocated memory 46.30 GiB is allocated by PyTorch, and 165.92 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":29683.796639992557,"meanTps":29681.610684764604,"stepMs":137.98773956298828,"jitter":0.00030961397417906086,"achievedTflops":133.56771524892469,"nominalPeakTflops":468.0,"mfuNominalPct":28.540110095924078,"configuredPeakTflops":468.0,"mfuConfiguredPct":28.540110095924078,"vramAllocatedGb":20.656325632,"vramAllocatedPct":20.256472891370752,"vramReservedGb":21.019754496,"vramReservedPct":20.612866717780665,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":32759.65312671036,"meanTps":32759.244340151028,"stepMs":250.0636978149414,"jitter":0.00015863506335010038,"achievedTflops":147.40809855120648,"nominalPeakTflops":468.0,"mfuNominalPct":31.497456955386,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.497456955386,"vramAllocatedGb":31.389135872,"vramAllocatedPct":30.781523839346942,"vramReservedGb":32.052871168,"vramReservedPct":31.43241094627953,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":31884.484136149433,"meanTps":31883.657599899943,"stepMs":513.8549499511719,"jitter":0.00023870583434528055,"achievedTflops":143.47011433902384,"nominalPeakTflops":468.0,"mfuNominalPct":30.656007337398258,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.656007337398258,"vramAllocatedGb":52.854756352,"vramAllocatedPct":51.83162573529932,"vramReservedGb":54.125395968,"vramReservedPct":53.07766907305211,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":29574.624665890267,"meanTps":29579.89648591019,"stepMs":138.49710845947266,"jitter":0.0005719492430298241,"achievedTflops":133.07647582537967,"nominalPeakTflops":468.0,"mfuNominalPct":28.43514440713241,"configuredPeakTflops":468.0,"mfuConfiguredPct":28.43514440713241,"vramAllocatedGb":20.675662848,"vramAllocatedPct":20.27543578916182,"vramReservedGb":22.468886528,"vramReservedPct":22.033947322588254,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":32597.272184847778,"meanTps":32596.5434547306,"stepMs":251.3093719482422,"jitter":0.0002794197456182102,"achievedTflops":146.67743556804442,"nominalPeakTflops":468.0,"mfuNominalPct":31.34133238633428,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.34133238633428,"vramAllocatedGb":31.406937088,"vramAllocatedPct":30.79898047010313,"vramReservedGb":32.329695232,"vramReservedPct":31.703876416373017,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":31788.90972934857,"meanTps":31788.98012631679,"stepMs":515.3998718261719,"jitter":0.00015906987798827647,"achievedTflops":143.04005967629033,"nominalPeakTflops":468.0,"mfuNominalPct":30.564115315446653,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.564115315446653,"vramAllocatedGb":52.872582144,"vramAllocatedPct":51.84910646632807,"vramReservedGb":54.377054208,"vramReservedPct":53.324455864046186,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":21395.712567440096,"meanTps":21387.03244603346,"stepMs":382.8804473876953,"jitter":0.0019506485236816023,"achievedTflops":101.72566549997116,"nominalPeakTflops":312.0,"mfuNominalPct":32.60437996793947,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.60437996793947,"vramAllocatedGb":31.393465344,"vramAllocatedPct":36.944685164090195,"vramReservedGb":32.006733824,"vramReservedPct":37.66639621657808,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":22738.249090452522,"meanTps":22736.90958122349,"stepMs":720.5480041503906,"jitter":0.0010085809543102454,"achievedTflops":108.10873971780714,"nominalPeakTflops":312.0,"mfuNominalPct":34.65023708904075,"configuredPeakTflops":312.0,"mfuConfiguredPct":34.65023708904075,"vramAllocatedGb":52.861645824,"vramAllocatedPct":62.20902473885627,"vramReservedGb":54.13797888,"vramReservedPct":63.711048246033485,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.426397696,"vramAllocatedPct":98.178457315225,"vramReservedGb":83.514884096,"vramReservedPct":98.28259052108432,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.14 GiB of which 866.75 MiB is free. Process 764857 has 78.28 GiB memory in use. Of the allocated memory 77.70 GiB is allocated by PyTorch, and 84.39 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":21381.216684624702,"meanTps":21378.540946984172,"stepMs":383.14002990722656,"jitter":0.003133897022497293,"achievedTflops":101.65674499443726,"nominalPeakTflops":312.0,"mfuNominalPct":32.582290062319636,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.582290062319636,"vramAllocatedGb":31.393465344,"vramAllocatedPct":36.944685164090195,"vramReservedGb":32.006733824,"vramReservedPct":37.66639621657808,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":22738.538006016533,"meanTps":22731.599417167305,"stepMs":720.5388488769531,"jitter":0.0014317610844307878,"achievedTflops":108.11011336348169,"nominalPeakTflops":312.0,"mfuNominalPct":34.650677360090285,"configuredPeakTflops":312.0,"mfuConfiguredPct":34.650677360090285,"vramAllocatedGb":52.861645824,"vramAllocatedPct":62.20902473885627,"vramReservedGb":54.13797888,"vramReservedPct":63.711048246033485,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.426397696,"vramAllocatedPct":98.178457315225,"vramReservedGb":83.514884096,"vramReservedPct":98.28259052108432,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.14 GiB of which 866.75 MiB is free. Process 852916 has 78.28 GiB memory in use. Of the allocated memory 77.70 GiB is allocated by PyTorch, and 84.39 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":81235.52045734036,"meanTps":81188.91663758366,"stepMs":403.37034606933594,"jitter":0.0008381775354063787,"achievedTflops":386.2333331835458,"nominalPeakTflops":2250.0,"mfuNominalPct":17.165925919268705,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.165925919268705,"vramAllocatedGb":95.798006784,"vramAllocatedPct":50.02428312356439,"vramReservedGb":98.434023424,"vramReservedPct":51.40077149889258,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":85578.63826154212,"meanTps":85560.71065916757,"stepMs":765.798583984375,"jitter":0.00046846583092927585,"achievedTflops":406.8826360560071,"nominalPeakTflops":2250.0,"mfuNominalPct":18.083672713600315,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.083672713600315,"vramAllocatedGb":181.670728192,"vramAllocatedPct":94.8657310045262,"vramReservedGb":186.837368832,"vramReservedPct":97.56367329841785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.176456704,"vramAllocatedPct":98.78510992208012,"vramReservedGb":190.266212352,"vramReservedPct":99.35416398594985,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.14 GiB is free. Including non-PyTorch memory, this process has 177.20 GiB memory in use. Of the allocated memory 176.18 GiB is allocated by PyTorch, and 199.27 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":81310.14598945584,"meanTps":81313.44242831724,"stepMs":403.00013732910156,"jitter":0.00042595742526688765,"achievedTflops":386.5881394043626,"nominalPeakTflops":2250.0,"mfuNominalPct":17.18169508463834,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.18169508463834,"vramAllocatedGb":95.798006784,"vramAllocatedPct":50.02428312356439,"vramReservedGb":98.434023424,"vramReservedPct":51.40077149889258,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":85731.62400007123,"meanTps":85709.18842902059,"stepMs":764.4320373535156,"jitter":0.0003845457780217094,"achievedTflops":407.6100049629703,"nominalPeakTflops":2250.0,"mfuNominalPct":18.11600022057646,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.11600022057646,"vramAllocatedGb":181.670728192,"vramAllocatedPct":94.8657310045262,"vramReservedGb":186.837368832,"vramReservedPct":97.56367329841785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.176456704,"vramAllocatedPct":98.78510992208012,"vramReservedGb":190.266212352,"vramReservedPct":99.35416398594985,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.14 GiB is free. Including non-PyTorch memory, this process has 177.20 GiB memory in use. Of the allocated memory 176.18 GiB is allocated by PyTorch, and 199.27 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":79126.09520008425,"meanTps":79059.90485842037,"stepMs":414.1238098144531,"jitter":0.0008400119778844656,"achievedTflops":376.20409543600863,"nominalPeakTflops":2250.0,"mfuNominalPct":16.72018201937816,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.72018201937816,"vramAllocatedGb":95.798006784,"vramAllocatedPct":33.329318375521254,"vramReservedGb":98.434023424,"vramReservedPct":34.246421359050196,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":85864.42314181976,"meanTps":85838.03970533813,"stepMs":763.249755859375,"jitter":0.0003279363497278387,"achievedTflops":408.24139693131974,"nominalPeakTflops":2250.0,"mfuNominalPct":18.144062085836435,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.144062085836435,"vramAllocatedGb":181.670728192,"vramAllocatedPct":63.20550648904776,"vramReservedGb":186.837368832,"vramReservedPct":65.0030450454682,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.274079232,"vramAllocatedPct":99.59831367781754,"vramReservedGb":286.506614784,"vramReservedPct":99.67921568931462,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 384.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 25.62 MiB is free. Including non-PyTorch memory, this process has 267.65 GiB memory in use. Of the allocated memory 266.61 GiB is allocated by PyTorch, and 221.76 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":79296.68229386315,"meanTps":79286.05368482404,"stepMs":413.23292541503906,"jitter":0.00030821767653375996,"achievedTflops":377.01514978092314,"nominalPeakTflops":2250.0,"mfuNominalPct":16.75622887915214,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.75622887915214,"vramAllocatedGb":95.798006784,"vramAllocatedPct":33.329318375521254,"vramReservedGb":98.434023424,"vramReservedPct":34.246421359050196,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":85917.37482193264,"meanTps":85843.91410103538,"stepMs":762.7793579101562,"jitter":0.0002086752604166458,"achievedTflops":408.49315507594076,"nominalPeakTflops":2250.0,"mfuNominalPct":18.155251336708478,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.155251336708478,"vramAllocatedGb":181.670728192,"vramAllocatedPct":63.20550648904776,"vramReservedGb":186.837368832,"vramReservedPct":65.0030450454682,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.274079232,"vramAllocatedPct":99.59831367781754,"vramReservedGb":286.506614784,"vramReservedPct":99.67921568931462,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 384.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 25.62 MiB is free. Including non-PyTorch memory, this process has 267.65 GiB memory in use. Of the allocated memory 266.61 GiB is allocated by PyTorch, and 221.76 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15083.167907707699,"meanTps":15083.858321625221,"stepMs":135.7804946899414,"jitter":0.0009689329804128455,"achievedTflops":71.7127456457951,"nominalPeakTflops":165.2,"mfuNominalPct":43.40965232796314,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":34.89388321099429,"vramAllocatedGb":15.292329984,"vramAllocatedPct":60.537195844889,"vramReservedGb":15.55038208,"vramReservedPct":61.55873738173007,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":16872.582472475442,"meanTps":16871.401336082487,"stepMs":242.76070404052734,"jitter":0.0007062810986258549,"achievedTflops":80.22049629362121,"nominalPeakTflops":165.2,"mfuNominalPct":48.55962245376587,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":39.03357211594577,"vramAllocatedGb":20.659375104,"vramAllocatedPct":81.78352402886993,"vramReservedGb":21.10783488,"vramReservedPct":83.5588255896309,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.649087488,"vramAllocatedPct":97.57745472534913,"vramReservedGb":24.733810688,"vramReservedPct":97.91284540527639,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 24.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 23.69 MiB is free. Including non-PyTorch memory, this process has 23.49 GiB memory in use. Of the allocated memory 22.96 GiB is allocated by PyTorch, and 80.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15067.876532928674,"meanTps":15060.088013689943,"stepMs":135.9182891845703,"jitter":0.0012038839675281876,"achievedTflops":71.64004298301147,"nominalPeakTflops":165.2,"mfuNominalPct":43.36564345218613,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":34.85850765534578,"vramAllocatedGb":15.292329984,"vramAllocatedPct":60.537195844889,"vramReservedGb":15.55038208,"vramReservedPct":61.55873738173007,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":16861.178345217304,"meanTps":16864.786244467025,"stepMs":242.92489624023438,"jitter":0.0012050069012405902,"achievedTflops":80.1662755037726,"nominalPeakTflops":165.2,"mfuNominalPct":48.52680115240472,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":39.00718944308131,"vramAllocatedGb":20.659375104,"vramAllocatedPct":81.78352402886993,"vramReservedGb":21.10783488,"vramReservedPct":83.5588255896309,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.649087488,"vramAllocatedPct":97.57745472534913,"vramReservedGb":24.733810688,"vramReservedPct":97.91284540527639,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 24.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 23.69 MiB is free. Including non-PyTorch memory, this process has 23.49 GiB memory in use. Of the allocated memory 22.96 GiB is allocated by PyTorch, and 80.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22678.1337195316,"meanTps":22672.36849280016,"stepMs":180.61450958251953,"jitter":0.00024123910675668853,"achievedTflops":107.82292188891115,"nominalPeakTflops":209.5,"mfuNominalPct":51.46678849112704,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":40.090796083541356,"vramAllocatedGb":20.659375104,"vramAllocatedPct":61.360486869970764,"vramReservedGb":21.021851648,"vramReservedPct":62.437079802081186,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":26070.210604230866,"meanTps":26070.646354164674,"stepMs":314.22837829589844,"jitter":0.00019611272050952953,"achievedTflops":123.9505118177558,"nominalPeakTflops":209.5,"mfuNominalPct":59.16492210871398,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":46.08736812805002,"vramAllocatedGb":31.393465344,"vramAllocatedPct":93.24184823336824,"vramReservedGb":32.052871168,"vramReservedPct":95.20035192488116,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.92795136,"vramAllocatedPct":97.79943085931569,"vramReservedGb":32.996589568,"vramReservedPct":98.00329345629942,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 48.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 31.88 MiB is free. Including non-PyTorch memory, this process has 31.32 GiB memory in use. Of the allocated memory 30.67 GiB is allocated by PyTorch, and 65.46 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22672.37308864369,"meanTps":22670.780931847872,"stepMs":180.660400390625,"jitter":0.00023141050156422088,"achievedTflops":107.79553303663873,"nominalPeakTflops":209.5,"mfuNominalPct":51.45371505328818,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":40.080612340862345,"vramAllocatedGb":20.659375104,"vramAllocatedPct":61.360486869970764,"vramReservedGb":21.021851648,"vramReservedPct":62.437079802081186,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":26065.231277738894,"meanTps":26064.840758255403,"stepMs":314.2884063720703,"jitter":0.00014604871874606285,"achievedTflops":123.92683766810825,"nominalPeakTflops":209.5,"mfuNominalPct":59.15362179861969,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":46.07856558876298,"vramAllocatedGb":31.393465344,"vramAllocatedPct":93.24184823336824,"vramReservedGb":32.052871168,"vramReservedPct":95.20035192488116,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.92795136,"vramAllocatedPct":97.79943085931569,"vramReservedGb":32.996589568,"vramReservedPct":98.00329345629942,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 48.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 31.88 MiB is free. Including non-PyTorch memory, this process has 31.32 GiB memory in use. Of the allocated memory 30.67 GiB is allocated by PyTorch, and 65.46 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":47009.76536530532,"meanTps":47001.356048145084,"stepMs":174.26166534423828,"jitter":0.0006207351879802941,"achievedTflops":223.50738035528468,"nominalPeakTflops":989.5,"mfuNominalPct":22.587911102100524,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.587911102100524,"vramAllocatedGb":31.443534848,"vramAllocatedPct":36.9847822513382,"vramReservedGb":32.069648384,"vramReservedPct":37.72123484502769,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":51500.362627666116,"meanTps":51488.74252350792,"stepMs":318.13368225097656,"jitter":0.00042325035928221543,"achievedTflops":244.85787258900305,"nominalPeakTflops":989.5,"mfuNominalPct":24.74561622930804,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.74561622930804,"vramAllocatedGb":52.911715328,"vramAllocatedPct":62.23626826343751,"vramReservedGb":54.20089344,"vramReservedPct":63.752636317665484,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.4764672,"vramAllocatedPct":98.1874009212851,"vramReservedGb":83.556827136,"vramReservedPct":98.28192256934595,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.18 GiB of which 642.19 MiB is free. Including non-PyTorch memory, this process has 78.54 GiB memory in use. Of the allocated memory 77.74 GiB is allocated by PyTorch, and 76.64 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":47000.439695470726,"meanTps":47004.35059384753,"stepMs":174.2962417602539,"jitter":0.0004022978235796446,"achievedTflops":223.4630415669799,"nominalPeakTflops":989.5,"mfuNominalPct":22.58343017351995,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.58343017351995,"vramAllocatedGb":31.443534848,"vramAllocatedPct":36.9847822513382,"vramReservedGb":32.069648384,"vramReservedPct":37.72123484502769,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":51477.65481216542,"meanTps":51475.55614686246,"stepMs":318.2740173339844,"jitter":0.00028644940557778,"achievedTflops":244.74990854543228,"nominalPeakTflops":989.5,"mfuNominalPct":24.73470525977082,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.73470525977082,"vramAllocatedGb":52.911715328,"vramAllocatedPct":62.23626826343751,"vramReservedGb":54.20089344,"vramReservedPct":63.752636317665484,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.4764672,"vramAllocatedPct":98.1874009212851,"vramReservedGb":83.556827136,"vramReservedPct":98.28192256934595,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.18 GiB of which 642.19 MiB is free. Including non-PyTorch memory, this process has 78.54 GiB memory in use. Of the allocated memory 77.74 GiB is allocated by PyTorch, and 76.64 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":56651.34148496287,"meanTps":56637.40827524964,"stepMs":289.20762634277344,"jitter":0.002288100215311031,"achievedTflops":269.34814140258743,"nominalPeakTflops":989.5,"mfuNominalPct":27.22063076327311,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.22063076327311,"vramAllocatedGb":52.911715328,"vramAllocatedPct":35.24610158923559,"vramReservedGb":54.167339008,"vramReservedPct":36.08251068141467,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":59370.206230582204,"meanTps":59375.1327275879,"stepMs":551.9266662597656,"jitter":0.0025448158923488734,"achievedTflops":282.27495207929434,"nominalPeakTflops":989.5,"mfuNominalPct":28.52702901256133,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.52702901256133,"vramAllocatedGb":95.848076288,"vramAllocatedPct":63.84731647873692,"vramReservedGb":98.343845888,"vramReservedPct":65.50982445678343,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.271254016,"vramAllocatedPct":98.76798832574964,"vramReservedGb":148.990066688,"vramReservedPct":99.24681129158789,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 769.44 MiB is free. Including non-PyTorch memory, this process has 139.05 GiB memory in use. Of the allocated memory 138.09 GiB is allocated by PyTorch, and 245.51 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":56507.03821241577,"meanTps":56479.639428080314,"stepMs":289.94618225097656,"jitter":0.0018350399457149392,"achievedTflops":268.6620531787245,"nominalPeakTflops":989.5,"mfuNominalPct":27.15129390386301,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.15129390386301,"vramAllocatedGb":52.911715328,"vramAllocatedPct":35.24610158923559,"vramReservedGb":54.167339008,"vramReservedPct":36.08251068141467,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":59303.657584729066,"meanTps":59315.25244098002,"stepMs":552.5460205078125,"jitter":0.0029476228427664283,"achievedTflops":281.95854731987373,"nominalPeakTflops":989.5,"mfuNominalPct":28.495052786242923,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.495052786242923,"vramAllocatedGb":95.848076288,"vramAllocatedPct":63.84731647873692,"vramReservedGb":98.343845888,"vramReservedPct":65.50982445678343,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.271254016,"vramAllocatedPct":98.76798832574964,"vramReservedGb":148.990066688,"vramReservedPct":99.24681129158789,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 769.44 MiB is free. Including non-PyTorch memory, this process has 139.05 GiB memory in use. Of the allocated memory 138.09 GiB is allocated by PyTorch, and 245.51 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":14782.87659690702,"meanTps":14794.895156684195,"stepMs":277.07733154296875,"jitter":0.00253344131191164,"achievedTflops":70.28501411599578,"nominalPeakTflops":364.2,"mfuNominalPct":19.298466259197085,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.569294035197649,"vramAllocatedGb":20.659375104,"vramAllocatedPct":40.60663810559834,"vramReservedGb":21.021851648,"vramReservedPct":41.31909691763309,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":15363.178489629305,"meanTps":15358.403251944623,"stepMs":533.2229919433594,"jitter":0.0017354903843726976,"achievedTflops":73.04405268701778,"nominalPeakTflops":364.2,"mfuNominalPct":20.05602764607847,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.18046674828553,"vramAllocatedGb":31.393465344,"vramAllocatedPct":61.70482309785024,"vramReservedGb":32.052871168,"vramReservedPct":63.00090555557703,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":49.718988288,"vramAllocatedPct":97.72420289694057,"vramReservedGb":49.973035008,"vramReservedPct":98.22353955010762,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.38 GiB of which 416.50 MiB is free. Process 2319972 has 46.97 GiB memory in use. Of the allocated memory 46.30 GiB is allocated by PyTorch, and 162.28 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":9948.090574665752,"meanTps":9948.449452463932,"stepMs":411.7373046875,"jitter":0.0010602630944644335,"achievedTflops":47.298080443550404,"nominalPeakTflops":154.8,"mfuNominalPct":30.554315532009305,"configuredPeakTflops":180.6,"mfuConfiguredPct":26.189413313150837,"vramAllocatedGb":20.659375104,"vramAllocatedPct":40.48164502527237,"vramReservedGb":21.021851648,"vramReservedPct":41.191910786474146,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":11326.737519470005,"meanTps":11325.307619741441,"stepMs":723.2444458007812,"jitter":0.0015001619036937588,"achievedTflops":53.85284124002592,"nominalPeakTflops":154.8,"mfuNominalPct":34.78865713179969,"configuredPeakTflops":180.6,"mfuConfiguredPct":29.81884897011402,"vramAllocatedGb":31.393465344,"vramAllocatedPct":61.5148867655217,"vramReservedGb":32.052871168,"vramReservedPct":62.80697969477961,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.718988288,"vramAllocatedPct":97.42339372601822,"vramReservedGb":49.952063488,"vramReservedPct":97.8801000621536,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 754.25 MiB is free. Process 1415479 has 46.79 GiB memory in use. Of the allocated memory 46.30 GiB is allocated by PyTorch, and 162.28 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":9992.536218903828,"meanTps":9992.380624961177,"stepMs":409.90594482421875,"jitter":0.0011884047764863647,"achievedTflops":47.509396739955214,"nominalPeakTflops":154.8,"mfuNominalPct":30.69082476741293,"configuredPeakTflops":180.6,"mfuConfiguredPct":26.306421229211086,"vramAllocatedGb":20.659375104,"vramAllocatedPct":40.48164502527237,"vramReservedGb":21.021851648,"vramReservedPct":41.191910786474146,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":11321.452102454212,"meanTps":11326.27647332514,"stepMs":723.5820922851562,"jitter":0.0009455739180099414,"achievedTflops":53.82771178655799,"nominalPeakTflops":154.8,"mfuNominalPct":34.77242363472738,"configuredPeakTflops":180.6,"mfuConfiguredPct":29.804934544052045,"vramAllocatedGb":31.393465344,"vramAllocatedPct":61.5148867655217,"vramReservedGb":32.052871168,"vramReservedPct":62.80697969477961,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.718988288,"vramAllocatedPct":97.42339372601822,"vramReservedGb":49.952063488,"vramReservedPct":97.8801000621536,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 754.25 MiB is free. Process 1449495 has 46.79 GiB memory in use. Of the allocated memory 46.30 GiB is allocated by PyTorch, and 162.28 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":29178.48056066589,"meanTps":29177.077330927652,"stepMs":140.37742614746094,"jitter":0.0003758160607188762,"achievedTflops":138.728745021033,"nominalPeakTflops":468.0,"mfuNominalPct":29.642894235263462,"configuredPeakTflops":468.0,"mfuConfiguredPct":29.642894235263462,"vramAllocatedGb":20.659375104,"vramAllocatedPct":20.259463333523993,"vramReservedGb":21.021851648,"vramReservedPct":20.614923274372284,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":32285.309595700055,"meanTps":32285.216500477243,"stepMs":253.7376937866211,"jitter":0.0001147905970150196,"achievedTflops":153.50012738033985,"nominalPeakTflops":468.0,"mfuNominalPct":32.79917251716664,"configuredPeakTflops":468.0,"mfuConfiguredPct":32.79917251716664,"vramAllocatedGb":31.393465344,"vramAllocatedPct":30.785769504029243,"vramReservedGb":32.052871168,"vramReservedPct":31.43241094627953,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":31511.599966825626,"meanTps":31509.15742759293,"stepMs":519.9355163574219,"jitter":0.0010801059114773648,"achievedTflops":149.82153398678486,"nominalPeakTflops":468.0,"mfuNominalPct":32.01314828777454,"configuredPeakTflops":468.0,"mfuConfiguredPct":32.01314828777454,"vramAllocatedGb":52.861645824,"vramAllocatedPct":51.83838184503975,"vramReservedGb":54.12749312,"vramReservedPct":53.07972562964372,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":29083.49848642769,"meanTps":29076.23857825878,"stepMs":140.83587646484375,"jitter":0.0003569232336058558,"achievedTflops":138.27715385845812,"nominalPeakTflops":468.0,"mfuNominalPct":29.546400397106435,"configuredPeakTflops":468.0,"mfuConfiguredPct":29.546400397106435,"vramAllocatedGb":20.67743232,"vramAllocatedPct":20.277171008785995,"vramReservedGb":22.47098368,"vramReservedPct":22.03600387917987,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":32143.280703967008,"meanTps":32144.949303698424,"stepMs":254.8588638305664,"jitter":0.00024441587109133605,"achievedTflops":152.82485267349253,"nominalPeakTflops":468.0,"mfuNominalPct":32.654883049891566,"configuredPeakTflops":468.0,"mfuConfiguredPct":32.654883049891566,"vramAllocatedGb":31.41280256,"vramAllocatedPct":30.80473240182031,"vramReservedGb":32.333889536,"vramReservedPct":31.707989529556254,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":31478.163794292275,"meanTps":31477.807219112397,"stepMs":520.48779296875,"jitter":0.00015708027935408398,"achievedTflops":149.6625620950095,"nominalPeakTflops":468.0,"mfuNominalPct":31.979179934831087,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.979179934831087,"vramAllocatedGb":52.880495616,"vramAllocatedPct":51.85686675409175,"vramReservedGb":54.38963712,"vramReservedPct":53.33679520359589,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15609.147403632789,"meanTps":15533.09283249915,"stepMs":131.20511627197266,"jitter":0.008659648606814655,"achievedTflops":82.16805440681549,"nominalPeakTflops":312.0,"mfuNominalPct":26.33591487397932,"configuredPeakTflops":312.0,"mfuConfiguredPct":26.33591487397932,"vramAllocatedGb":15.296172544,"vramAllocatedPct":18.000952512293654,"vramReservedGb":15.365832704,"vramReservedPct":18.082930486100615,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":18449.192521448822,"meanTps":18454.163355576417,"stepMs":222.01513671875,"jitter":0.002973966404361682,"achievedTflops":97.11832527838193,"nominalPeakTflops":312.0,"mfuNominalPct":31.127668358455747,"configuredPeakTflops":312.0,"mfuConfiguredPct":31.127668358455747,"vramAllocatedGb":20.663554048,"vramAllocatedPct":24.31743327183936,"vramReservedGb":21.049114624,"vramReservedPct":24.77117146021453,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":20433.832434561376,"meanTps":20428.596531508963,"stepMs":400.90374755859375,"jitter":0.002308688156808797,"achievedTflops":107.56566081450592,"nominalPeakTflops":312.0,"mfuNominalPct":34.476173337982665,"configuredPeakTflops":312.0,"mfuConfiguredPct":34.476173337982665,"vramAllocatedGb":31.398284288,"vramAllocatedPct":36.95035622865578,"vramReservedGb":32.05496832,"vramReservedPct":37.72315988536207,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21750.215083159277,"meanTps":21740.999008040006,"stepMs":753.2799072265625,"jitter":0.0010445212011468725,"achievedTflops":114.49522578645349,"nominalPeakTflops":312.0,"mfuNominalPct":36.69718775206842,"configuredPeakTflops":312.0,"mfuConfiguredPct":36.69718775206842,"vramAllocatedGb":52.867744768,"vramAllocatedPct":62.21620214228862,"vramReservedGb":54.112813056,"vramReservedPct":63.68143241884184,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.42993664,"vramAllocatedPct":98.18262204092383,"vramReservedGb":83.55053568,"vramReservedPct":98.32454627627247,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.14 GiB of which 832.75 MiB is free. Process 703773 has 78.32 GiB memory in use. Of the allocated memory 77.70 GiB is allocated by PyTorch, and 115.01 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15619.932288041688,"meanTps":15530.052157594608,"stepMs":131.1145248413086,"jitter":0.005749164953152502,"achievedTflops":82.22482707645376,"nominalPeakTflops":312.0,"mfuNominalPct":26.35411124245313,"configuredPeakTflops":312.0,"mfuConfiguredPct":26.35411124245313,"vramAllocatedGb":15.296172544,"vramAllocatedPct":18.000952512293654,"vramReservedGb":15.365832704,"vramReservedPct":18.082930486100615,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":18437.226051642483,"meanTps":18438.80955934102,"stepMs":222.15923309326172,"jitter":0.0039016519708621783,"achievedTflops":97.05533262947687,"nominalPeakTflops":312.0,"mfuNominalPct":31.10747840688361,"configuredPeakTflops":312.0,"mfuConfiguredPct":31.10747840688361,"vramAllocatedGb":20.663554048,"vramAllocatedPct":24.31743327183936,"vramReservedGb":21.049114624,"vramReservedPct":24.77117146021453,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":20425.187717226734,"meanTps":20423.32632899681,"stepMs":401.07342529296875,"jitter":0.003414214889943019,"achievedTflops":107.52015419035045,"nominalPeakTflops":312.0,"mfuNominalPct":34.46158788152258,"configuredPeakTflops":312.0,"mfuConfiguredPct":34.46158788152258,"vramAllocatedGb":31.398284288,"vramAllocatedPct":36.95035622865578,"vramReservedGb":32.05496832,"vramReservedPct":37.72315988536207,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21728.588147244678,"meanTps":21730.91954674901,"stepMs":754.0296630859375,"jitter":0.0016320691917174827,"achievedTflops":114.38137951407668,"nominalPeakTflops":312.0,"mfuNominalPct":36.66069856220407,"configuredPeakTflops":312.0,"mfuConfiguredPct":36.66069856220407,"vramAllocatedGb":52.867744768,"vramAllocatedPct":62.21620214228862,"vramReservedGb":54.112813056,"vramReservedPct":63.68143241884184,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.42993664,"vramAllocatedPct":98.18262204092383,"vramReservedGb":83.55053568,"vramReservedPct":98.32454627627247,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.14 GiB of which 832.75 MiB is free. Process 736416 has 78.32 GiB memory in use. Of the allocated memory 77.70 GiB is allocated by PyTorch, and 115.01 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":12005.544719890479,"meanTps":11949.20807958197,"stepMs":170.5878448486328,"jitter":0.012814354257223732,"achievedTflops":63.19834301121604,"nominalPeakTflops":2250.0,"mfuNominalPct":2.808815244942935,"configuredPeakTflops":2250.0,"mfuConfiguredPct":2.808815244942935,"vramAllocatedGb":15.296172544,"vramAllocatedPct":7.987432011745646,"vramReservedGb":15.365832704,"vramReservedPct":8.023807503086816,"warmupSteps":17,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":23609.830999597834,"meanTps":23279.367269546197,"stepMs":173.48705291748047,"jitter":0.050452538383445926,"achievedTflops":124.2844229697757,"nominalPeakTflops":2250.0,"mfuNominalPct":5.523752131990031,"configuredPeakTflops":2250.0,"mfuConfiguredPct":5.523752131990031,"vramAllocatedGb":20.663554048,"vramAllocatedPct":10.790198175698059,"vramReservedGb":21.049114624,"vramReservedPct":10.991532128904378,"warmupSteps":18,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":47063.58157481488,"meanTps":46940.958653109534,"stepMs":174.0624008178711,"jitter":0.024937307842699884,"achievedTflops":247.7472235619336,"nominalPeakTflops":2250.0,"mfuNominalPct":11.010987713863717,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.010987713863717,"vramAllocatedGb":31.398284288,"vramAllocatedPct":16.395713392644478,"vramReservedGb":32.05496832,"vramReservedPct":16.73862395041381,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":68682.06973354032,"meanTps":68652.36262567226,"stepMs":238.54843139648438,"jitter":0.0021665472628987345,"achievedTflops":361.54902613866057,"nominalPeakTflops":2250.0,"mfuNominalPct":16.068845606162693,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.068845606162693,"vramAllocatedGb":52.867744768,"vramAllocatedPct":27.60674382653732,"vramReservedGb":54.112813056,"vramReservedPct":28.256899822867357,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":75227.20797039651,"meanTps":75219.09283661215,"stepMs":435.5870819091797,"jitter":0.001087489345066415,"achievedTflops":396.0032638262976,"nominalPeakTflops":2250.0,"mfuNominalPct":17.600145058946563,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.600145058946563,"vramAllocatedGb":95.806665728,"vramAllocatedPct":50.02880469432299,"vramReservedGb":98.20962816,"vramReservedPct":51.283595655732995,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":79244.31923685166,"meanTps":79232.35377354844,"stepMs":827.011962890625,"jitter":0.00042224745759409537,"achievedTflops":417.1497242039796,"nominalPeakTflops":2250.0,"mfuNominalPct":18.53998774239909,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.53998774239909,"vramAllocatedGb":181.684507648,"vramAllocatedPct":94.87292642989435,"vramReservedGb":186.925449216,"vramReservedPct":97.60966755461133,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.179995648,"vramAllocatedPct":98.7869579055879,"vramReservedGb":190.35848704,"vramReservedPct":99.40234844481921,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.07 GiB is free. Including non-PyTorch memory, this process has 177.27 GiB memory in use. Of the allocated memory 176.19 GiB is allocated by PyTorch, and 263.90 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11600.09002278572,"meanTps":11688.133242028154,"stepMs":176.55035400390625,"jitter":0.02188776805027724,"achievedTflops":61.06399045820927,"nominalPeakTflops":2250.0,"mfuNominalPct":2.7139551314759673,"configuredPeakTflops":2250.0,"mfuConfiguredPct":2.7139551314759673,"vramAllocatedGb":15.296172544,"vramAllocatedPct":7.987432011745646,"vramReservedGb":15.365832704,"vramReservedPct":8.023807503086816,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":23341.04245378297,"meanTps":23220.49727124911,"stepMs":175.48487854003906,"jitter":0.014969576804711624,"achievedTflops":122.869494192096,"nominalPeakTflops":2250.0,"mfuNominalPct":5.4608664085375995,"configuredPeakTflops":2250.0,"mfuConfiguredPct":5.4608664085375995,"vramAllocatedGb":20.663554048,"vramAllocatedPct":10.790198175698059,"vramReservedGb":21.049114624,"vramReservedPct":10.991532128904378,"warmupSteps":26,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":46697.21607669857,"meanTps":46613.95635388835,"stepMs":175.42801666259766,"jitter":0.016739528638734583,"achievedTflops":245.81864031497193,"nominalPeakTflops":2250.0,"mfuNominalPct":10.925272902887642,"configuredPeakTflops":2250.0,"mfuConfiguredPct":10.925272902887642,"vramAllocatedGb":31.398284288,"vramAllocatedPct":16.395713392644478,"vramReservedGb":32.05496832,"vramReservedPct":16.73862395041381,"warmupSteps":15,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":68393.27691675728,"meanTps":68261.81978859329,"stepMs":239.5557098388672,"jitter":0.004725671385047813,"achievedTflops":360.02879295307326,"nominalPeakTflops":2250.0,"mfuNominalPct":16.001279686803258,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.001279686803258,"vramAllocatedGb":52.867744768,"vramAllocatedPct":27.60674382653732,"vramReservedGb":54.112813056,"vramReservedPct":28.256899822867357,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":75310.67945128519,"meanTps":75169.0487070757,"stepMs":435.1042938232422,"jitter":0.0010443453909074858,"achievedTflops":396.4426657363265,"nominalPeakTflops":2250.0,"mfuNominalPct":17.61967403272562,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.61967403272562,"vramAllocatedGb":95.806665728,"vramAllocatedPct":50.02880469432299,"vramReservedGb":98.20962816,"vramReservedPct":51.283595655732995,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":79339.38763048334,"meanTps":79293.14154740694,"stepMs":826.02099609375,"jitter":0.0002434204296174656,"achievedTflops":417.65017337896006,"nominalPeakTflops":2250.0,"mfuNominalPct":18.56222992795378,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.56222992795378,"vramAllocatedGb":181.684507648,"vramAllocatedPct":94.87292642989435,"vramReservedGb":186.925449216,"vramReservedPct":97.60966755461133,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.179995648,"vramAllocatedPct":98.7869579055879,"vramReservedGb":190.35848704,"vramReservedPct":99.40234844481921,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.07 GiB is free. Including non-PyTorch memory, this process has 177.27 GiB memory in use. Of the allocated memory 176.19 GiB is allocated by PyTorch, and 263.90 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":21220.04988740788,"meanTps":21204.95909950421,"stepMs":96.51249694824219,"jitter":0.0024323264758832467,"achievedTflops":111.70438516443704,"nominalPeakTflops":2250.0,"mfuNominalPct":4.9646393406416465,"configuredPeakTflops":2250.0,"mfuConfiguredPct":4.9646393406416465,"vramAllocatedGb":15.296172544,"vramAllocatedPct":5.3217287265211715,"vramReservedGb":15.365832704,"vramReservedPct":5.345964362821671,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":37810.227097259914,"meanTps":37810.91679115753,"stepMs":108.33047866821289,"jitter":0.0011456949449410235,"achievedTflops":199.03667490119565,"nominalPeakTflops":2250.0,"mfuNominalPct":8.84607444005314,"configuredPeakTflops":2250.0,"mfuConfiguredPct":8.84607444005314,"vramAllocatedGb":20.663554048,"vramAllocatedPct":7.189107526928303,"vramReservedGb":21.049114624,"vramReservedPct":7.3232488480471005,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":52945.951128801484,"meanTps":52918.8825986032,"stepMs":154.72382354736328,"jitter":0.001141706544701935,"achievedTflops":278.7125831074828,"nominalPeakTflops":2250.0,"mfuNominalPct":12.387225915888125,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.387225915888125,"vramAllocatedGb":31.398284288,"vramAllocatedPct":10.9238537273477,"vramReservedGb":32.05496832,"vramReservedPct":11.152322271834207,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":64761.91643414524,"meanTps":64594.61490131418,"stepMs":252.98818969726562,"jitter":0.0005985900944484728,"achievedTflops":340.91296183236886,"nominalPeakTflops":2250.0,"mfuNominalPct":15.151687192549728,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.151687192549728,"vramAllocatedGb":52.867744768,"vramAllocatedPct":18.393346128186497,"vramReservedGb":54.112813056,"vramReservedPct":18.82652087537704,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":73800.30033084781,"meanTps":73793.64564418771,"stepMs":444.009033203125,"jitter":0.0004930013619910472,"achievedTflops":388.4918846632915,"nominalPeakTflops":2250.0,"mfuNominalPct":17.266305985035178,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.266305985035178,"vramAllocatedGb":95.806665728,"vramAllocatedPct":33.332330929864085,"vramReservedGb":98.20962816,"vramReservedPct":34.168351455020996,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":79622.6597989823,"meanTps":79622.64946633787,"stepMs":823.082275390625,"jitter":0.00014995991214188995,"achievedTflops":419.1413453405845,"nominalPeakTflops":2250.0,"mfuNominalPct":18.628504237359312,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.628504237359312,"vramAllocatedGb":181.684507648,"vramAllocatedPct":63.21030053321927,"vramReservedGb":186.925449216,"vramReservedPct":65.03368930686283,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.447145984,"vramAllocatedPct":99.31061331302772,"vramReservedGb":286.410145792,"vramReservedPct":99.64565292683477,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 768.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 697.62 MiB is free. Including non-PyTorch memory, this process has 266.99 GiB memory in use. Of the allocated memory 265.84 GiB is allocated by PyTorch, and 336.39 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":21314.28534676835,"meanTps":21256.135923586684,"stepMs":96.0857925415039,"jitter":0.005601707180560001,"achievedTflops":112.20044969323892,"nominalPeakTflops":2250.0,"mfuNominalPct":4.986686653032841,"configuredPeakTflops":2250.0,"mfuConfiguredPct":4.986686653032841,"vramAllocatedGb":15.296172544,"vramAllocatedPct":5.3217287265211715,"vramReservedGb":15.365832704,"vramReservedPct":5.345964362821671,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":37936.48082801968,"meanTps":37939.58704673734,"stepMs":107.96995162963867,"jitter":0.0015159842882593868,"achievedTflops":199.70128669259418,"nominalPeakTflops":2250.0,"mfuNominalPct":8.875612741893075,"configuredPeakTflops":2250.0,"mfuConfiguredPct":8.875612741893075,"vramAllocatedGb":20.663554048,"vramAllocatedPct":7.189107526928303,"vramReservedGb":21.049114624,"vramReservedPct":7.3232488480471005,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":53007.411615669815,"meanTps":52996.766449549126,"stepMs":154.54442596435547,"jitter":0.0008337245223705815,"achievedTflops":279.03611702629837,"nominalPeakTflops":2250.0,"mfuNominalPct":12.401605201168817,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.401605201168817,"vramAllocatedGb":31.398284288,"vramAllocatedPct":10.9238537273477,"vramReservedGb":32.05496832,"vramReservedPct":11.152322271834207,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":64838.85762090636,"meanTps":64855.69017353764,"stepMs":252.68798065185547,"jitter":0.0005897217475471779,"achievedTflops":341.31798764553025,"nominalPeakTflops":2250.0,"mfuNominalPct":15.169688339801343,"configuredPeakTflops":2250.0,"mfuConfiguredPct":15.169688339801343,"vramAllocatedGb":52.867744768,"vramAllocatedPct":18.393346128186497,"vramReservedGb":54.112813056,"vramReservedPct":18.82652087537704,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":73703.24537680646,"meanTps":73683.57173847368,"stepMs":444.5937194824219,"jitter":0.0003354269401536073,"achievedTflops":387.9809780430962,"nominalPeakTflops":2250.0,"mfuNominalPct":17.24359902413761,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.24359902413761,"vramAllocatedGb":95.806665728,"vramAllocatedPct":33.332330929864085,"vramReservedGb":98.20962816,"vramReservedPct":34.168351455020996,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":79593.87460149944,"meanTps":79550.22246967912,"stepMs":823.3799438476562,"jitter":0.00030717480030217366,"achievedTflops":418.9898172902867,"nominalPeakTflops":2250.0,"mfuNominalPct":18.621769657346075,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.621769657346075,"vramAllocatedGb":181.684507648,"vramAllocatedPct":63.21030053321927,"vramReservedGb":186.925449216,"vramReservedPct":65.03368930686283,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.447145984,"vramAllocatedPct":99.31061331302772,"vramReservedGb":286.410145792,"vramReservedPct":99.64565292683477,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 768.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 697.62 MiB is free. Including non-PyTorch memory, this process has 266.99 GiB memory in use. Of the allocated memory 265.84 GiB is allocated by PyTorch, and 336.39 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":14568.374532309219,"meanTps":14572.011623064252,"stepMs":140.57848358154297,"jitter":0.001278715035024335,"achievedTflops":76.6893258315347,"nominalPeakTflops":165.2,"mfuNominalPct":46.42211006751496,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":37.31538034135174,"vramAllocatedGb":15.296172544,"vramAllocatedPct":60.552407248743684,"vramReservedGb":15.365832704,"vramReservedPct":60.828168414826195,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":16307.074382421522,"meanTps":16305.628287437108,"stepMs":251.17932891845703,"jitter":0.000751194251895895,"achievedTflops":85.84200920281866,"nominalPeakTflops":165.2,"mfuNominalPct":51.962475304369654,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":41.768879670498094,"vramAllocatedGb":20.663554048,"vramAllocatedPct":81.80006706394813,"vramReservedGb":21.049114624,"vramReservedPct":83.32637182743422,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.682068992,"vramAllocatedPct":97.70801741716889,"vramReservedGb":24.754782208,"vramReservedPct":97.99586460606092,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 3.69 MiB is free. Including non-PyTorch memory, this process has 23.51 GiB memory in use. Of the allocated memory 22.99 GiB is allocated by PyTorch, and 69.41 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":14586.200093080704,"meanTps":14585.461205887814,"stepMs":140.40668487548828,"jitter":0.000986472705972298,"achievedTflops":76.78316129925298,"nominalPeakTflops":165.2,"mfuNominalPct":46.47891119809503,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":37.36103866641131,"vramAllocatedGb":15.296172544,"vramAllocatedPct":60.552407248743684,"vramReservedGb":15.365832704,"vramReservedPct":60.828168414826195,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":16296.313816319634,"meanTps":16296.09982818324,"stepMs":251.34518432617188,"jitter":0.0008668518478271031,"achievedTflops":85.78536454708926,"nominalPeakTflops":165.2,"mfuNominalPct":51.92818677184581,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":41.741317596507635,"vramAllocatedGb":20.663554048,"vramAllocatedPct":81.80006706394813,"vramReservedGb":21.049114624,"vramReservedPct":83.32637182743422,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.682068992,"vramAllocatedPct":97.70801741716889,"vramReservedGb":24.754782208,"vramReservedPct":97.99586460606092,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 3.69 MiB is free. Including non-PyTorch memory, this process has 23.51 GiB memory in use. Of the allocated memory 22.99 GiB is allocated by PyTorch, and 69.41 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":18053.797784626197,"meanTps":18052.296830338877,"stepMs":113.43873596191406,"jitter":0.00047154376304170886,"achievedTflops":95.03692932463184,"nominalPeakTflops":209.5,"mfuNominalPct":45.363689415098726,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":35.336699165741955,"vramAllocatedGb":15.296172544,"vramAllocatedPct":45.4312189900262,"vramReservedGb":15.365832704,"vramReservedPct":45.63811689044781,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":21925.41605822961,"meanTps":21914.959878837053,"stepMs":186.81515502929688,"jitter":0.0010483938405233045,"achievedTflops":115.41750058336876,"nominalPeakTflops":209.5,"mfuNominalPct":55.09188571998509,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":42.914617776053106,"vramAllocatedGb":20.663554048,"vramAllocatedPct":61.37289876709502,"vramReservedGb":21.049114624,"vramReservedPct":62.518053668544376,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25133.225813413348,"meanTps":25133.95873512569,"stepMs":325.9430389404297,"jitter":0.0002368161832101236,"achievedTflops":132.30371990559178,"nominalPeakTflops":209.5,"mfuNominalPct":63.152133606487716,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":49.19326394524788,"vramAllocatedGb":31.398284288,"vramAllocatedPct":93.25616099687394,"vramReservedGb":32.05496832,"vramReservedPct":95.20658068383987,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.931490304,"vramAllocatedPct":97.80994189005851,"vramReservedGb":32.986103808,"vramReservedPct":97.97214966150588,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 48.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 39.88 MiB is free. Including non-PyTorch memory, this process has 31.31 GiB memory in use. Of the allocated memory 30.67 GiB is allocated by PyTorch, and 52.08 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":18004.61297207239,"meanTps":17995.008157620385,"stepMs":113.74862670898438,"jitter":0.00045922110151677545,"achievedTflops":94.7780157370153,"nominalPeakTflops":209.5,"mfuNominalPct":45.24010297709561,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":35.240429730054736,"vramAllocatedGb":15.296172544,"vramAllocatedPct":45.4312189900262,"vramReservedGb":15.365832704,"vramReservedPct":45.63811689044781,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":21860.06859930078,"meanTps":21858.22449160894,"stepMs":187.3736114501953,"jitter":0.0005073917012908359,"achievedTflops":115.07350526948241,"nominalPeakTflops":209.5,"mfuNominalPct":54.92768747946654,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":42.78671319193397,"vramAllocatedGb":20.663554048,"vramAllocatedPct":61.37289876709502,"vramReservedGb":21.049114624,"vramReservedPct":62.518053668544376,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25125.121695433423,"meanTps":25122.334452054514,"stepMs":326.0481719970703,"jitter":0.00031098120905041966,"achievedTflops":132.26105904847546,"nominalPeakTflops":209.5,"mfuNominalPct":63.13177042886657,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":49.17740175478376,"vramAllocatedGb":31.398284288,"vramAllocatedPct":93.25616099687394,"vramReservedGb":32.05496832,"vramReservedPct":95.20658068383987,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.931490304,"vramAllocatedPct":97.80994189005851,"vramReservedGb":32.986103808,"vramReservedPct":97.97214966150588,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 48.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 39.88 MiB is free. Including non-PyTorch memory, this process has 31.31 GiB memory in use. Of the allocated memory 30.67 GiB is allocated by PyTorch, and 52.08 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":29482.021417550863,"meanTps":29476.60915043491,"stepMs":69.466064453125,"jitter":0.0010616685938010706,"achievedTflops":155.1961985634413,"nominalPeakTflops":989.5,"mfuNominalPct":15.68430505946855,"configuredPeakTflops":989.5,"mfuConfiguredPct":15.68430505946855,"vramAllocatedGb":15.346242048,"vramAllocatedPct":18.050687470900296,"vramReservedGb":15.470690304,"vramReservedPct":18.19706705805449,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":38465.17237364609,"meanTps":38469.29138875221,"stepMs":106.4859390258789,"jitter":0.0011319207369606997,"achievedTflops":202.48436987850502,"nominalPeakTflops":989.5,"mfuNominalPct":20.46330165523042,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.46330165523042,"vramAllocatedGb":20.713623552,"vramAllocatedPct":24.363954638377386,"vramReservedGb":21.153972224,"vramReservedPct":24.881905302236092,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":44716.796697924176,"meanTps":44715.21662183555,"stepMs":183.19738006591797,"jitter":0.00032773718958654377,"achievedTflops":235.3935220778558,"nominalPeakTflops":989.5,"mfuNominalPct":23.789138158449298,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.789138158449298,"vramAllocatedGb":31.448353792,"vramAllocatedPct":36.99045043067564,"vramReservedGb":32.11788288,"vramReservedPct":37.777969634553955,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":48894.550079739136,"meanTps":48756.465406285366,"stepMs":335.0884704589844,"jitter":0.0005864816138243284,"achievedTflops":257.3856180135593,"nominalPeakTflops":989.5,"mfuNominalPct":26.01168448848502,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.01168448848502,"vramAllocatedGb":52.917814272,"vramAllocatedPct":62.243442015272144,"vramReservedGb":54.175727616,"vramReservedPct":63.72303555791265,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.480006144,"vramAllocatedPct":98.19156352812534,"vramReservedGb":83.86510848,"vramReservedPct":98.64453187631815,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.18 GiB of which 606.19 MiB is free. Including non-PyTorch memory, this process has 78.58 GiB memory in use. Of the allocated memory 77.75 GiB is allocated by PyTorch, and 107.26 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":29419.639563859942,"meanTps":29414.942999063423,"stepMs":69.61336135864258,"jitter":0.0012376715852751355,"achievedTflops":154.8678144809846,"nominalPeakTflops":989.5,"mfuNominalPct":15.65111818908384,"configuredPeakTflops":989.5,"mfuConfiguredPct":15.65111818908384,"vramAllocatedGb":15.346242048,"vramAllocatedPct":18.050687470900296,"vramReservedGb":15.470690304,"vramReservedPct":18.19706705805449,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":38411.22123465033,"meanTps":38409.81036702394,"stepMs":106.63550567626953,"jitter":0.00040913051786984104,"achievedTflops":202.2003658897105,"nominalPeakTflops":989.5,"mfuNominalPct":20.434599887792874,"configuredPeakTflops":989.5,"mfuConfiguredPct":20.434599887792874,"vramAllocatedGb":20.713623552,"vramAllocatedPct":24.363954638377386,"vramReservedGb":21.153972224,"vramReservedPct":24.881905302236092,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":44672.76400972469,"meanTps":44675.298725453846,"stepMs":183.3779525756836,"jitter":0.000744207900617893,"achievedTflops":235.1617297687633,"nominalPeakTflops":989.5,"mfuNominalPct":23.765712962987703,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.765712962987703,"vramAllocatedGb":31.448353792,"vramAllocatedPct":36.99045043067564,"vramReservedGb":32.11788288,"vramReservedPct":37.777969634553955,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":48889.77472194312,"meanTps":48888.868065904724,"stepMs":335.12120056152344,"jitter":0.00022891273367156738,"achievedTflops":257.36048007046423,"nominalPeakTflops":989.5,"mfuNominalPct":26.00914401924853,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.00914401924853,"vramAllocatedGb":52.917814272,"vramAllocatedPct":62.243442015272144,"vramReservedGb":54.175727616,"vramReservedPct":63.72303555791265,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.480006144,"vramAllocatedPct":98.19156352812534,"vramReservedGb":83.86510848,"vramReservedPct":98.64453187631815,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.18 GiB of which 606.19 MiB is free. Including non-PyTorch memory, this process has 78.58 GiB memory in use. Of the allocated memory 77.75 GiB is allocated by PyTorch, and 107.26 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":30508.6539254928,"meanTps":30559.3049700979,"stepMs":67.12849426269531,"jitter":0.0012272902619556191,"achievedTflops":160.60049090478634,"nominalPeakTflops":989.5,"mfuNominalPct":16.23046901513758,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.23046901513758,"vramAllocatedGb":15.346242048,"vramAllocatedPct":10.222598206914947,"vramReservedGb":15.470690304,"vramReservedPct":10.305496972271323,"warmupSteps":17,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":41156.358710629975,"meanTps":41155.61713330261,"stepMs":99.52289581298828,"jitter":0.0005060518227183296,"achievedTflops":216.65103379922024,"nominalPeakTflops":989.5,"mfuNominalPct":21.895000889259247,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.895000889259247,"vramAllocatedGb":20.713623552,"vramAllocatedPct":13.797974143707865,"vramReservedGb":21.153972224,"vramReservedPct":14.091303776508179,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":48869.79067959178,"meanTps":48862.09425199557,"stepMs":167.62911987304688,"jitter":0.0006758633872887238,"achievedTflops":257.2552821479428,"nominalPeakTflops":989.5,"mfuNominalPct":25.99851259706345,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.99851259706345,"vramAllocatedGb":31.448353792,"vramAllocatedPct":20.948704189533064,"vramReservedGb":32.11788288,"vramReservedPct":21.39469786232009,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":53359.11141780375,"meanTps":53356.95676852523,"stepMs":307.0515899658203,"jitter":0.001555250588967324,"achievedTflops":280.8874986379476,"nominalPeakTflops":989.5,"mfuNominalPct":28.386811383319618,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.386811383319618,"vramAllocatedGb":52.917814272,"vramAllocatedPct":35.250164281183466,"vramReservedGb":54.175727616,"vramReservedPct":36.08809858813679,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":55915.82322310268,"meanTps":55893.29397277214,"stepMs":586.0237426757812,"jitter":0.003654766032874536,"achievedTflops":294.3462756798927,"nominalPeakTflops":989.5,"mfuNominalPct":29.746970760979558,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.746970760979558,"vramAllocatedGb":95.856735232,"vramAllocatedPct":63.85308446448427,"vramReservedGb":98.52420096,"vramReservedPct":65.62996445130904,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.27479296,"vramAllocatedPct":98.77034572389803,"vramReservedGb":149.006843904,"vramReservedPct":99.25798710503213,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 751.44 MiB is free. Including non-PyTorch memory, this process has 139.07 GiB memory in use. Of the allocated memory 138.09 GiB is allocated by PyTorch, and 258.14 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":31145.2116704165,"meanTps":31123.22765021179,"stepMs":65.75649642944336,"jitter":0.006243501027487994,"achievedTflops":163.951392146567,"nominalPeakTflops":989.5,"mfuNominalPct":16.569114921330673,"configuredPeakTflops":989.5,"mfuConfiguredPct":16.569114921330673,"vramAllocatedGb":15.346242048,"vramAllocatedPct":10.222598206914947,"vramReservedGb":15.470690304,"vramReservedPct":10.305496972271323,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":41355.07741767956,"meanTps":41358.31055097381,"stepMs":99.04467010498047,"jitter":0.0011700244099842065,"achievedTflops":217.69710820099712,"nominalPeakTflops":989.5,"mfuNominalPct":22.00071836291027,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.00071836291027,"vramAllocatedGb":20.713623552,"vramAllocatedPct":13.797974143707865,"vramReservedGb":21.153972224,"vramReservedPct":14.091303776508179,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":49009.89105868379,"meanTps":49008.89180742961,"stepMs":167.14993286132812,"jitter":0.0004832444735592419,"achievedTflops":257.99278402898517,"nominalPeakTflops":989.5,"mfuNominalPct":26.07304537938203,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.07304537938203,"vramAllocatedGb":31.448353792,"vramAllocatedPct":20.948704189533064,"vramReservedGb":32.11788288,"vramReservedPct":21.39469786232009,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":53489.586251724126,"meanTps":53503.14391352096,"stepMs":306.3026123046875,"jitter":0.0015193921564017022,"achievedTflops":281.57433072269004,"nominalPeakTflops":989.5,"mfuNominalPct":28.45622341815968,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.45622341815968,"vramAllocatedGb":52.917814272,"vramAllocatedPct":35.250164281183466,"vramReservedGb":54.175727616,"vramReservedPct":36.08809858813679,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":55927.607066699034,"meanTps":55931.55573586677,"stepMs":585.9002685546875,"jitter":0.0025356387749638255,"achievedTflops":294.4083069668495,"nominalPeakTflops":989.5,"mfuNominalPct":29.75323971367858,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.75323971367858,"vramAllocatedGb":95.856735232,"vramAllocatedPct":63.85308446448427,"vramReservedGb":98.52420096,"vramReservedPct":65.62996445130904,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.27479296,"vramAllocatedPct":98.77034572389803,"vramReservedGb":149.006843904,"vramReservedPct":99.25798710503213,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 751.44 MiB is free. Including non-PyTorch memory, this process has 139.07 GiB memory in use. Of the allocated memory 138.09 GiB is allocated by PyTorch, and 258.14 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":12788.362640732015,"meanTps":12818.769731892975,"stepMs":160.14559936523438,"jitter":0.004399783788975296,"achievedTflops":67.31917189744767,"nominalPeakTflops":364.2,"mfuNominalPct":18.48412188287965,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.912310891018846,"vramAllocatedGb":15.296172544,"vramAllocatedPct":30.06509827950881,"vramReservedGb":15.365832704,"vramReservedPct":30.202017469622668,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":14387.113593018948,"meanTps":14382.373850665783,"stepMs":284.69921875,"jitter":0.003972130889576444,"achievedTflops":75.73515079965775,"nominalPeakTflops":364.2,"mfuNominalPct":20.794934321707235,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.77658952524207,"vramAllocatedGb":20.663554048,"vramAllocatedPct":40.61485194874788,"vramReservedGb":21.049114624,"vramReservedPct":41.372683136700246,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14880.523955865006,"meanTps":14884.465999154381,"stepMs":550.5182495117188,"jitter":0.0042267875584651405,"achievedTflops":78.33251044338728,"nominalPeakTflops":364.2,"mfuNominalPct":21.508102812572016,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.35194767970786,"vramAllocatedGb":31.398284288,"vramAllocatedPct":61.7142948807127,"vramReservedGb":32.05496832,"vramReservedPct":63.00502757242835,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":49.723551232,"vramAllocatedPct":97.73317150391784,"vramReservedGb":49.918509056,"vramReservedPct":98.1163671119733,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.38 GiB of which 468.50 MiB is free. Process 2308332 has 46.92 GiB memory in use. Of the allocated memory 46.31 GiB is allocated by PyTorch, and 105.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":12815.706505231738,"meanTps":12836.006888410224,"stepMs":159.8039093017578,"jitter":0.005505331078499775,"achievedTflops":67.46311263218523,"nominalPeakTflops":364.2,"mfuNominalPct":18.52364432514696,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.944196146374601,"vramAllocatedGb":15.296172544,"vramAllocatedPct":30.06509827950881,"vramReservedGb":15.365832704,"vramReservedPct":30.202017469622668,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":14429.333475959169,"meanTps":14430.36994258955,"stepMs":283.86619567871094,"jitter":0.004201115276705321,"achievedTflops":75.95740032737213,"nominalPeakTflops":364.2,"mfuNominalPct":20.85595835457774,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.8258214744661,"vramAllocatedGb":20.663554048,"vramAllocatedPct":40.61485194874788,"vramReservedGb":21.049114624,"vramReservedPct":41.372683136700246,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14858.055730386484,"meanTps":14869.54599733507,"stepMs":551.3507385253906,"jitter":0.003546177890867604,"achievedTflops":78.21423554176683,"nominalPeakTflops":364.2,"mfuNominalPct":21.47562755128139,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.325747831227016,"vramAllocatedGb":31.398284288,"vramAllocatedPct":61.7142948807127,"vramReservedGb":32.05496832,"vramReservedPct":63.00502757242835,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":49.723551232,"vramAllocatedPct":97.73317150391784,"vramReservedGb":49.918509056,"vramReservedPct":98.1163671119733,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.38 GiB of which 468.50 MiB is free. Process 2312978 has 46.92 GiB memory in use. Of the allocated memory 46.31 GiB is allocated by PyTorch, and 105.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":7808.0111247717105,"meanTps":7808.276037736985,"stepMs":262.2947082519531,"jitter":0.003950553452040536,"achievedTflops":41.102122128717106,"nominalPeakTflops":154.8,"mfuNominalPct":26.551758481083404,"configuredPeakTflops":180.6,"mfuConfiguredPct":22.75865012664292,"vramAllocatedGb":15.296172544,"vramAllocatedPct":29.972553576978,"vramReservedGb":15.365832704,"vramReservedPct":30.109051310105354,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":9675.066872228674,"meanTps":9671.231853118148,"stepMs":423.3562469482422,"jitter":0.001158182935169575,"achievedTflops":50.930483298648596,"nominalPeakTflops":154.8,"mfuNominalPct":32.900829004294955,"configuredPeakTflops":180.6,"mfuConfiguredPct":28.200710575109966,"vramAllocatedGb":20.663554048,"vramAllocatedPct":40.489833585029714,"vramReservedGb":21.049114624,"vramReservedPct":41.24533205944144,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10986.076210052226,"meanTps":10986.616724491381,"stepMs":745.671142578125,"jitter":0.0009483265635738043,"achievedTflops":57.8317626454666,"nominalPeakTflops":154.8,"mfuNominalPct":37.359019796813044,"configuredPeakTflops":180.6,"mfuConfiguredPct":32.022016968696896,"vramAllocatedGb":31.398284288,"vramAllocatedPct":61.52432939287237,"vramReservedGb":32.05496832,"vramReservedPct":62.81108902346941,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.723551232,"vramAllocatedPct":97.43233472633155,"vramReservedGb":49.918509056,"vramReservedPct":97.81435080311692,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 806.25 MiB is free. Process 1348507 has 46.73 GiB memory in use. Of the allocated memory 46.31 GiB is allocated by PyTorch, and 105.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":7832.737240773268,"meanTps":7828.667133676556,"stepMs":261.4667053222656,"jitter":0.0019331660352477976,"achievedTflops":41.23228278338633,"nominalPeakTflops":154.8,"mfuNominalPct":26.63584159133484,"configuredPeakTflops":180.6,"mfuConfiguredPct":22.830721364001295,"vramAllocatedGb":15.296172544,"vramAllocatedPct":29.972553576978,"vramReservedGb":15.365832704,"vramReservedPct":30.109051310105354,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":9673.516045663198,"meanTps":9668.411594058118,"stepMs":423.4241180419922,"jitter":0.0011451211539744305,"achievedTflops":50.92231959833157,"nominalPeakTflops":154.8,"mfuNominalPct":32.89555529607982,"configuredPeakTflops":180.6,"mfuConfiguredPct":28.196190253782706,"vramAllocatedGb":20.663554048,"vramAllocatedPct":40.489833585029714,"vramReservedGb":21.049114624,"vramReservedPct":41.24533205944144,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10983.961156612193,"meanTps":10981.981835918405,"stepMs":745.8147277832031,"jitter":0.001071841322202589,"achievedTflops":57.82062880057168,"nominalPeakTflops":154.8,"mfuNominalPct":37.351827390550184,"configuredPeakTflops":180.6,"mfuConfiguredPct":32.01585204904302,"vramAllocatedGb":31.398284288,"vramAllocatedPct":61.52432939287237,"vramReservedGb":32.05496832,"vramReservedPct":62.81108902346941,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.723551232,"vramAllocatedPct":97.43233472633155,"vramReservedGb":49.918509056,"vramReservedPct":97.81435080311692,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 806.25 MiB is free. Process 1380474 has 46.73 GiB memory in use. Of the allocated memory 46.31 GiB is allocated by PyTorch, and 105.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":21125.895002036563,"meanTps":21126.525233109838,"stepMs":96.9426383972168,"jitter":0.0006129823513863867,"achievedTflops":111.20874478486978,"nominalPeakTflops":468.0,"mfuNominalPct":23.762552304459355,"configuredPeakTflops":468.0,"mfuConfiguredPct":23.762552304459355,"vramAllocatedGb":15.296172544,"vramAllocatedPct":15.000078426303617,"vramReservedGb":15.365832704,"vramReservedPct":15.0683901467803,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":28248.21343652321,"meanTps":28246.970779679836,"stepMs":145.0003204345703,"jitter":0.0005316318207557167,"achievedTflops":148.70131458989073,"nominalPeakTflops":468.0,"mfuNominalPct":31.773785168780073,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.773785168780073,"vramAllocatedGb":20.663554048,"vramAllocatedPct":20.26356138403688,"vramReservedGb":21.049114624,"vramReservedPct":20.64165851006331,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":31305.914071135157,"meanTps":31304.662555296476,"stepMs":261.67579650878906,"jitter":0.0007803255723231162,"achievedTflops":164.79734505252028,"nominalPeakTflops":468.0,"mfuNominalPct":35.21310791720519,"configuredPeakTflops":468.0,"mfuConfiguredPct":35.21310791720519,"vramAllocatedGb":31.398284288,"vramAllocatedPct":30.79049516580666,"vramReservedGb":32.05496832,"vramReservedPct":31.434467502871147,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":30670.275927470495,"meanTps":30667.018719582087,"stepMs":534.197998046875,"jitter":0.00014881392319733097,"achievedTflops":161.45128468028437,"nominalPeakTflops":468.0,"mfuNominalPct":34.498137752197515,"configuredPeakTflops":468.0,"mfuConfiguredPct":34.498137752197515,"vramAllocatedGb":52.867744768,"vramAllocatedPct":51.84436272934623,"vramReservedGb":54.112813056,"vramReservedPct":53.0653297335024,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30766.629264093877,"meanTps":30766.665585085375,"stepMs":1065.050048828125,"jitter":0.0001237073246768727,"achievedTflops":161.95849791885658,"nominalPeakTflops":468.0,"mfuNominalPct":34.60651664932833,"configuredPeakTflops":468.0,"mfuConfiguredPct":34.60651664932833,"vramAllocatedGb":95.806665728,"vramAllocatedPct":93.95209785642535,"vramReservedGb":98.20962816,"vramReservedPct":96.30854518544035,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":21116.36340551884,"meanTps":21107.131423049253,"stepMs":96.98639678955078,"jitter":0.000347493459236524,"achievedTflops":111.15856954332715,"nominalPeakTflops":468.0,"mfuNominalPct":23.751831099001524,"configuredPeakTflops":468.0,"mfuConfiguredPct":23.751831099001524,"vramAllocatedGb":15.296172544,"vramAllocatedPct":15.000078426303617,"vramReservedGb":15.365832704,"vramReservedPct":15.0683901467803,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":28247.65905032949,"meanTps":28237.144400062607,"stepMs":145.00316619873047,"jitter":0.0004502595226576231,"achievedTflops":148.698396247604,"nominalPeakTflops":468.0,"mfuNominalPct":31.773161591368375,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.773161591368375,"vramAllocatedGb":20.663554048,"vramAllocatedPct":20.26356138403688,"vramReservedGb":21.049114624,"vramReservedPct":20.64165851006331,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":31323.331767179126,"meanTps":31322.92806959988,"stepMs":261.53028869628906,"jitter":0.00016123821438817094,"achievedTflops":164.8890334810535,"nominalPeakTflops":468.0,"mfuNominalPct":35.232699461763566,"configuredPeakTflops":468.0,"mfuConfiguredPct":35.232699461763566,"vramAllocatedGb":31.398284288,"vramAllocatedPct":30.79049516580666,"vramReservedGb":32.05496832,"vramReservedPct":31.434467502871147,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":30691.73345486447,"meanTps":30691.40763297733,"stepMs":533.8245239257812,"jitter":0.00010620051197202527,"achievedTflops":161.564239169902,"nominalPeakTflops":468.0,"mfuNominalPct":34.52227332690214,"configuredPeakTflops":468.0,"mfuConfiguredPct":34.52227332690214,"vramAllocatedGb":52.867744768,"vramAllocatedPct":51.84436272934623,"vramReservedGb":54.112813056,"vramReservedPct":53.0653297335024,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30774.780715657605,"meanTps":30775.14906148308,"stepMs":1064.7679443359375,"jitter":0.0001978481822864545,"achievedTflops":162.00140794451423,"nominalPeakTflops":468.0,"mfuNominalPct":34.615685458229535,"configuredPeakTflops":468.0,"mfuConfiguredPct":34.615685458229535,"vramAllocatedGb":95.806665728,"vramAllocatedPct":93.95209785642535,"vramReservedGb":98.20962816,"vramReservedPct":96.30854518544035,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":18935.63454142808,"meanTps":18904.60821745834,"stepMs":432.62347412109375,"jitter":0.0029299122064112217,"achievedTflops":118.97850216553078,"nominalPeakTflops":312.0,"mfuNominalPct":38.134135309464995,"configuredPeakTflops":312.0,"mfuConfiguredPct":38.134135309464995,"vramAllocatedGb":31.406002176,"vramAllocatedPct":36.959438849486816,"vramReservedGb":32.017219584,"vramReservedPct":37.67873614457459,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19992.86757048212,"meanTps":19992.813220177337,"stepMs":819.4922485351562,"jitter":0.0017478291053085,"achievedTflops":125.6214272791082,"nominalPeakTflops":312.0,"mfuNominalPct":40.26327797407315,"configuredPeakTflops":312.0,"mfuConfiguredPct":40.26327797407315,"vramAllocatedGb":52.876102656,"vramAllocatedPct":62.22603793255304,"vramReservedGb":54.167339008,"vramReservedPct":63.74560004442374,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.437014528,"vramAllocatedPct":98.19095149232147,"vramReservedGb":83.525369856,"vramReservedPct":98.29493044908082,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.14 GiB of which 856.75 MiB is free. Process 764857 has 78.29 GiB memory in use. Of the allocated memory 77.71 GiB is allocated by PyTorch, and 84.26 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":18894.723447947003,"meanTps":18845.86228610758,"stepMs":433.56019592285156,"jitter":0.005173691397360037,"achievedTflops":118.72144499569137,"nominalPeakTflops":312.0,"mfuNominalPct":38.05174519092672,"configuredPeakTflops":312.0,"mfuConfiguredPct":38.05174519092672,"vramAllocatedGb":31.406002176,"vramAllocatedPct":36.959438849486816,"vramReservedGb":32.017219584,"vramReservedPct":37.67873614457459,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19995.433538937305,"meanTps":19987.986379450384,"stepMs":819.3870849609375,"jitter":0.0013749417311404426,"achievedTflops":125.63755005982271,"nominalPeakTflops":312.0,"mfuNominalPct":40.26844553199446,"configuredPeakTflops":312.0,"mfuConfiguredPct":40.26844553199446,"vramAllocatedGb":52.876102656,"vramAllocatedPct":62.22603793255304,"vramReservedGb":54.167339008,"vramReservedPct":63.74560004442374,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.437014528,"vramAllocatedPct":98.19095149232147,"vramReservedGb":83.525369856,"vramReservedPct":98.29493044908082,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.14 GiB of which 856.75 MiB is free. Process 852916 has 78.29 GiB memory in use. Of the allocated memory 77.71 GiB is allocated by PyTorch, and 84.26 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":66146.27854969045,"meanTps":66128.25899978509,"stepMs":495.38690185546875,"jitter":0.0009432455064234609,"achievedTflops":415.61771423333624,"nominalPeakTflops":2250.0,"mfuNominalPct":18.4718984103705,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.4718984103705,"vramAllocatedGb":95.816303616,"vramAllocatedPct":50.03383745496396,"vramReservedGb":98.465480704,"vramReservedPct":51.41719801896168,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":68925.02549844165,"meanTps":68889.27546305269,"stepMs":950.8302612304688,"jitter":0.0006527725210344089,"achievedTflops":433.0774486370677,"nominalPeakTflops":2250.0,"mfuNominalPct":19.247886606091896,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.247886606091896,"vramAllocatedGb":181.696705536,"vramAllocatedPct":94.87929598416073,"vramReservedGb":186.868826112,"vramReservedPct":97.58009981848696,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.187073536,"vramAllocatedPct":98.79065387260344,"vramReservedGb":190.278795264,"vramReservedPct":99.36073459397748,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.13 GiB is free. Including non-PyTorch memory, this process has 177.21 GiB memory in use. Of the allocated memory 176.19 GiB is allocated by PyTorch, and 199.15 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":66158.71323511517,"meanTps":66130.24324169711,"stepMs":495.2937927246094,"jitter":0.0004269268043998359,"achievedTflops":415.69584524307317,"nominalPeakTflops":2250.0,"mfuNominalPct":18.47537089969214,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.47537089969214,"vramAllocatedGb":95.816303616,"vramAllocatedPct":50.03383745496396,"vramReservedGb":98.465480704,"vramReservedPct":51.41719801896168,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":69028.8474634076,"meanTps":69007.59570381242,"stepMs":949.4001770019531,"jitter":0.0002519137722679998,"achievedTflops":433.729794448691,"nominalPeakTflops":2250.0,"mfuNominalPct":19.276879753275157,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.276879753275157,"vramAllocatedGb":181.696705536,"vramAllocatedPct":94.87929598416073,"vramReservedGb":186.868826112,"vramReservedPct":97.58009981848696,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.187073536,"vramAllocatedPct":98.79065387260344,"vramReservedGb":190.278795264,"vramReservedPct":99.36073459397748,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.13 GiB is free. Including non-PyTorch memory, this process has 177.21 GiB memory in use. Of the allocated memory 176.19 GiB is allocated by PyTorch, and 199.15 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":64676.57215111309,"meanTps":64681.78765389034,"stepMs":506.64404296875,"jitter":0.00024250512838675913,"achievedTflops":406.38309019455596,"nominalPeakTflops":2250.0,"mfuNominalPct":18.0614706753136,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.0614706753136,"vramAllocatedGb":95.816303616,"vramAllocatedPct":33.335684071003485,"vramReservedGb":98.465480704,"vramReservedPct":34.2573657381197,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":69311.7428058059,"meanTps":69299.63576769926,"stepMs":945.5252075195312,"jitter":0.00044875693969945423,"achievedTflops":435.50731418453694,"nominalPeakTflops":2250.0,"mfuNominalPct":19.355880630423865,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.355880630423865,"vramAllocatedGb":181.696705536,"vramAllocatedPct":63.214544330207424,"vramReservedGb":186.868826112,"vramReservedPct":65.01398942453771,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.284696064,"vramAllocatedPct":99.60200740575351,"vramReservedGb":286.519197696,"vramReservedPct":99.68359344094242,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 384.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 13.62 MiB is free. Including non-PyTorch memory, this process has 267.66 GiB memory in use. Of the allocated memory 266.62 GiB is allocated by PyTorch, and 223.64 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":64762.736718275395,"meanTps":64763.561124762266,"stepMs":505.969970703125,"jitter":0.00026835847423319124,"achievedTflops":406.92448906441075,"nominalPeakTflops":2250.0,"mfuNominalPct":18.085532847307142,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.085532847307142,"vramAllocatedGb":95.816303616,"vramAllocatedPct":33.335684071003485,"vramReservedGb":98.465480704,"vramReservedPct":34.2573657381197,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":69360.1312991631,"meanTps":69360.08179041882,"stepMs":944.8655700683594,"jitter":0.00013738350275690453,"achievedTflops":435.8113541917039,"nominalPeakTflops":2250.0,"mfuNominalPct":19.369393519631288,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.369393519631288,"vramAllocatedGb":181.696705536,"vramAllocatedPct":63.214544330207424,"vramReservedGb":186.868826112,"vramReservedPct":65.01398942453771,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.284696064,"vramAllocatedPct":99.60200740575351,"vramReservedGb":286.519197696,"vramReservedPct":99.68359344094242,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 384.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 13.62 MiB is free. Including non-PyTorch memory, this process has 267.66 GiB memory in use. Of the allocated memory 266.62 GiB is allocated by PyTorch, and 223.64 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":15201.685737146716,"meanTps":15201.21028689497,"stepMs":269.4438018798828,"jitter":0.0003741287212240339,"achievedTflops":95.51693635825858,"nominalPeakTflops":165.2,"mfuNominalPct":57.818968739865966,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":46.47649161864824,"vramAllocatedGb":20.670919168,"vramAllocatedPct":81.8292230991846,"vramReservedGb":20.759707648,"vramReservedPct":82.18070685660769,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.65970432,"vramAllocatedPct":97.6194831957463,"vramReservedGb":24.744296448,"vramReservedPct":97.95435500566866,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 24.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 13.69 MiB is free. Including non-PyTorch memory, this process has 23.50 GiB memory in use. Of the allocated memory 22.97 GiB is allocated by PyTorch, and 80.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":15202.012018423493,"meanTps":15203.790885092647,"stepMs":269.4380187988281,"jitter":0.0006363515346397559,"achievedTflops":95.51898648536209,"nominalPeakTflops":165.2,"mfuNominalPct":57.820209736901994,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":46.47748916650495,"vramAllocatedGb":20.670919168,"vramAllocatedPct":81.8292230991846,"vramReservedGb":20.759707648,"vramReservedPct":82.18070685660769,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.65970432,"vramAllocatedPct":97.6194831957463,"vramReservedGb":24.744296448,"vramReservedPct":97.95435500566866,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 24.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 13.69 MiB is free. Including non-PyTorch memory, this process has 23.50 GiB memory in use. Of the allocated memory 22.97 GiB is allocated by PyTorch, and 80.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":20441.945062218423,"meanTps":20442.047727011464,"stepMs":200.37232208251953,"jitter":0.0004250241773651277,"achievedTflops":128.44312132934672,"nominalPeakTflops":209.5,"mfuNominalPct":61.309365789664305,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":47.75781341609125,"vramAllocatedGb":20.670919168,"vramAllocatedPct":61.39477393741265,"vramReservedGb":20.759707648,"vramReservedPct":61.65848493224278,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":23399.9677377639,"meanTps":23400.118583006082,"stepMs":350.08595275878906,"jitter":0.00019541197180013818,"achievedTflops":147.02930108150062,"nominalPeakTflops":209.5,"mfuNominalPct":70.18105063556115,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":54.66854009054877,"vramAllocatedGb":31.406002176,"vramAllocatedPct":93.27908392474102,"vramReservedGb":32.063356928,"vramReservedPct":95.2314957196747,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.938568192,"vramAllocatedPct":97.83096395154415,"vramReservedGb":33.00917248,"vramReservedPct":98.04066601005167,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 48.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 19.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.68 GiB is allocated by PyTorch, and 67.33 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":20444.40260648459,"meanTps":20442.371684392936,"stepMs":200.34823608398438,"jitter":0.000320796614049847,"achievedTflops":128.45856284703942,"nominalPeakTflops":209.5,"mfuNominalPct":61.31673644250091,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":47.76355489226527,"vramAllocatedGb":20.670919168,"vramAllocatedPct":61.39477393741265,"vramReservedGb":20.759707648,"vramReservedPct":61.65848493224278,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":23400.09726669092,"meanTps":23399.095341700307,"stepMs":350.0840148925781,"jitter":0.0002659788783611077,"achievedTflops":147.03011495217868,"nominalPeakTflops":209.5,"mfuNominalPct":70.18143911798505,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":54.66884270452751,"vramAllocatedGb":31.406002176,"vramAllocatedPct":93.27908392474102,"vramReservedGb":32.063356928,"vramReservedPct":95.2314957196747,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.938568192,"vramAllocatedPct":97.83096395154415,"vramReservedGb":33.00917248,"vramReservedPct":98.04066601005167,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 48.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 19.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.68 GiB is allocated by PyTorch, and 67.33 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":40699.77784480157,"meanTps":40687.317029542006,"stepMs":201.27873992919922,"jitter":0.0002721077740597976,"achievedTflops":255.7294077391474,"nominalPeakTflops":989.5,"mfuNominalPct":25.844305986775886,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.844305986775886,"vramAllocatedGb":31.45607168,"vramAllocatedPct":36.99952843060472,"vramReservedGb":32.080134144,"vramReservedPct":37.733568494924704,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":44101.96921155963,"meanTps":44100.1892135336,"stepMs":371.502685546875,"jitter":0.002144846024939622,"achievedTflops":277.1064380156752,"nominalPeakTflops":989.5,"mfuNominalPct":28.00469307889593,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.00469307889593,"vramAllocatedGb":52.92617216,"vramAllocatedPct":62.25327280144982,"vramReservedGb":54.230253568,"vramReservedPct":63.78717053737713,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.487084032,"vramAllocatedPct":98.19988874180584,"vramReservedGb":83.567312896,"vramReservedPct":98.29425621924297,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.18 GiB of which 632.19 MiB is free. Including non-PyTorch memory, this process has 78.55 GiB memory in use. Of the allocated memory 77.75 GiB is allocated by PyTorch, and 76.51 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":40694.596542644125,"meanTps":40692.223672192304,"stepMs":201.3043670654297,"jitter":0.0003500089936216916,"achievedTflops":255.6968520004627,"nominalPeakTflops":989.5,"mfuNominalPct":25.841015866646053,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.841015866646053,"vramAllocatedGb":31.45607168,"vramAllocatedPct":36.99952843060472,"vramReservedGb":32.080134144,"vramReservedPct":37.733568494924704,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":44338.08827077579,"meanTps":44327.57707581948,"stepMs":369.52427673339844,"jitter":0.0002224774573945407,"achievedTflops":278.59004776410006,"nominalPeakTflops":989.5,"mfuNominalPct":28.154628374340582,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.154628374340582,"vramAllocatedGb":52.92617216,"vramAllocatedPct":62.25327280144982,"vramReservedGb":54.230253568,"vramReservedPct":63.78717053737713,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.487084032,"vramAllocatedPct":98.19988874180584,"vramReservedGb":83.567312896,"vramReservedPct":98.29425621924297,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.18 GiB of which 632.19 MiB is free. Including non-PyTorch memory, this process has 78.55 GiB memory in use. Of the allocated memory 77.75 GiB is allocated by PyTorch, and 76.51 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":47959.19718796839,"meanTps":47975.75292569427,"stepMs":341.6237335205078,"jitter":0.0028133348799501247,"achievedTflops":301.34260534030534,"nominalPeakTflops":989.5,"mfuNominalPct":30.454027826205696,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.454027826205696,"vramAllocatedGb":52.92617216,"vramAllocatedPct":35.25573172437999,"vramReservedGb":54.17992192,"vramReservedPct":36.090892541497844,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":50175.93951569791,"meanTps":50174.988280458725,"stepMs":653.06201171875,"jitter":0.0013881316679015844,"achievedTflops":315.27108929277915,"nominalPeakTflops":989.5,"mfuNominalPct":31.861656320644684,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.861656320644684,"vramAllocatedGb":95.86637312,"vramAllocatedPct":63.85950455458053,"vramReservedGb":98.371108864,"vramReservedPct":65.52798515363033,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.281870848,"vramAllocatedPct":98.77506052019483,"vramReservedGb":148.998455296,"vramReservedPct":99.25239919831,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 761.44 MiB is free. Including non-PyTorch memory, this process has 139.06 GiB memory in use. Of the allocated memory 138.10 GiB is allocated by PyTorch, and 243.39 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":47969.331559977225,"meanTps":47960.29826329022,"stepMs":341.5515594482422,"jitter":0.002371214375603418,"achievedTflops":301.40628276285804,"nominalPeakTflops":989.5,"mfuNominalPct":30.460463139247906,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.460463139247906,"vramAllocatedGb":52.92617216,"vramAllocatedPct":35.25573172437999,"vramReservedGb":54.17992192,"vramReservedPct":36.090892541497844,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":50168.41409041625,"meanTps":50184.41758300021,"stepMs":653.1599731445312,"jitter":0.0015116423472016159,"achievedTflops":315.2238046968385,"nominalPeakTflops":989.5,"mfuNominalPct":31.856877685380343,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.856877685380343,"vramAllocatedGb":95.86637312,"vramAllocatedPct":63.85950455458053,"vramReservedGb":98.371108864,"vramReservedPct":65.52798515363033,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.281870848,"vramAllocatedPct":98.77506052019483,"vramReservedGb":148.998455296,"vramReservedPct":99.25239919831,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 761.44 MiB is free. Including non-PyTorch memory, this process has 139.06 GiB memory in use. Of the allocated memory 138.10 GiB is allocated by PyTorch, and 243.39 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":13732.034384892257,"meanTps":13732.63853859825,"stepMs":298.2806396484375,"jitter":0.004185277960233516,"achievedTflops":86.28265819270653,"nominalPeakTflops":364.2,"mfuNominalPct":23.69100993759103,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":19.113037002790833,"vramAllocatedGb":20.670919168,"vramAllocatedPct":40.62932831896424,"vramReservedGb":20.759707648,"vramReservedPct":40.80384481121807,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14068.15529737064,"meanTps":14061.555359126047,"stepMs":582.3080444335938,"jitter":0.0049827484936641395,"achievedTflops":88.39461079855639,"nominalPeakTflops":364.2,"mfuNominalPct":24.270898077582753,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":19.5808694635571,"vramAllocatedGb":31.406002176,"vramAllocatedPct":61.72946462729883,"vramReservedGb":32.063356928,"vramReservedPct":63.02151563983362,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":49.73114112,"vramAllocatedPct":97.7480896625614,"vramReservedGb":49.98561792,"vramReservedPct":98.24827165121555,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.38 GiB of which 404.50 MiB is free. Process 2319972 has 46.98 GiB memory in use. Of the allocated memory 46.32 GiB is allocated by PyTorch, and 162.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":9147.201109786502,"meanTps":9149.084070827694,"stepMs":447.78724670410156,"jitter":0.001247263615355795,"achievedTflops":57.47471967037659,"nominalPeakTflops":154.8,"mfuNominalPct":37.12837188008823,"configuredPeakTflops":180.6,"mfuConfiguredPct":31.824318754361347,"vramAllocatedGb":20.670919168,"vramAllocatedPct":40.50426539489364,"vramReservedGb":20.759707648,"vramReservedPct":40.678244700250154,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10317.424766543365,"meanTps":10317.640853546553,"stepMs":793.99658203125,"jitter":0.0006648361498796202,"achievedTflops":64.8276001653495,"nominalPeakTflops":154.8,"mfuNominalPct":41.878294680458325,"configuredPeakTflops":180.6,"mfuConfiguredPct":35.89568115467857,"vramAllocatedGb":31.406002176,"vramAllocatedPct":61.53945244479374,"vramReservedGb":32.084328448,"vramReservedPct":62.86861962512649,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.73114112,"vramAllocatedPct":97.44720696473425,"vramReservedGb":49.9646464,"vramReservedPct":97.90475603429235,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 742.25 MiB is free. Process 1415479 has 46.80 GiB memory in use. Of the allocated memory 46.32 GiB is allocated by PyTorch, and 162.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":9123.092496584815,"meanTps":9123.816448668487,"stepMs":448.97056579589844,"jitter":0.0013771011056210234,"achievedTflops":57.32323772865716,"nominalPeakTflops":154.8,"mfuNominalPct":37.03051532858989,"configuredPeakTflops":180.6,"mfuConfiguredPct":31.740441710219912,"vramAllocatedGb":20.670919168,"vramAllocatedPct":40.50426539489364,"vramReservedGb":20.759707648,"vramReservedPct":40.678244700250154,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10306.88521493295,"meanTps":10310.391715288715,"stepMs":794.8085021972656,"jitter":0.0010344613875942457,"achievedTflops":64.7613768728921,"nominalPeakTflops":154.8,"mfuNominalPct":41.835514775770086,"configuredPeakTflops":180.6,"mfuConfiguredPct":35.859012664945794,"vramAllocatedGb":31.406002176,"vramAllocatedPct":61.53945244479374,"vramReservedGb":32.084328448,"vramReservedPct":62.86861962512649,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.73114112,"vramAllocatedPct":97.44720696473425,"vramReservedGb":49.9646464,"vramReservedPct":97.90475603429235,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 742.25 MiB is free. Process 1449495 has 46.80 GiB memory in use. Of the allocated memory 46.32 GiB is allocated by PyTorch, and 162.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":26626.144461776817,"meanTps":26555.411550827117,"stepMs":153.8337631225586,"jitter":0.006443401846096712,"achievedTflops":167.3003764185623,"nominalPeakTflops":468.0,"mfuNominalPct":35.747943679179976,"configuredPeakTflops":468.0,"mfuConfiguredPct":35.747943679179976,"vramAllocatedGb":20.670919168,"vramAllocatedPct":20.270783934469108,"vramReservedGb":20.759707648,"vramReservedPct":20.357853700420115,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":29543.24418244726,"meanTps":29527.16170044945,"stepMs":277.28843688964844,"jitter":0.0012153773075891292,"achievedTflops":185.6294244720363,"nominalPeakTflops":468.0,"mfuNominalPct":39.66440693846929,"configuredPeakTflops":468.0,"mfuConfiguredPct":39.66440693846929,"vramAllocatedGb":31.406002176,"vramAllocatedPct":30.7980636555679,"vramReservedGb":32.063356928,"vramReservedPct":31.442693729237615,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":29092.465949024463,"meanTps":29094.292427246037,"stepMs":563.1698608398438,"jitter":0.0002873780518533972,"achievedTflops":182.79704413093245,"nominalPeakTflops":468.0,"mfuNominalPct":39.059197463874455,"configuredPeakTflops":468.0,"mfuConfiguredPct":39.059197463874455,"vramAllocatedGb":52.876102656,"vramAllocatedPct":51.852558830372,"vramReservedGb":54.13797888,"vramReservedPct":53.090008412601804,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":26672.799672893972,"meanTps":26677.293507182047,"stepMs":153.56468200683594,"jitter":0.0004717088756326376,"achievedTflops":167.59352567240913,"nominalPeakTflops":468.0,"mfuNominalPct":35.81058240863443,"configuredPeakTflops":468.0,"mfuConfiguredPct":35.81058240863443,"vramAllocatedGb":20.688016384,"vramAllocatedPct":20.287550192834313,"vramReservedGb":21.718106112,"vramReservedPct":21.297700062789243,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":29450.035738010487,"meanTps":29442.36436497987,"stepMs":278.1660461425781,"jitter":0.0008326357866119872,"achievedTflops":185.04376672267472,"nominalPeakTflops":468.0,"mfuNominalPct":39.53926639373391,"configuredPeakTflops":468.0,"mfuConfiguredPct":39.53926639373391,"vramAllocatedGb":31.423419392,"vramAllocatedPct":30.815143719565373,"vramReservedGb":33.401339904,"vramReservedPct":32.75477683468948,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":29101.35218311266,"meanTps":29098.919480531702,"stepMs":562.9978942871094,"jitter":0.00026978551699708076,"achievedTflops":182.85287911335107,"nominalPeakTflops":468.0,"mfuNominalPct":39.07112801567331,"configuredPeakTflops":468.0,"mfuConfiguredPct":39.07112801567331,"vramAllocatedGb":52.895732736,"vramAllocatedPct":51.871808923077715,"vramReservedGb":54.404317184,"vramReservedPct":53.351191099737214,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":16462.665395660006,"meanTps":16427.32823125115,"stepMs":497.6107940673828,"jitter":0.0014853995414673895,"achievedTflops":136.99808487312012,"nominalPeakTflops":312.0,"mfuNominalPct":43.9096425875385,"configuredPeakTflops":312.0,"mfuConfiguredPct":43.9096425875385,"vramAllocatedGb":31.420412416,"vramAllocatedPct":36.97639721244883,"vramReservedGb":31.505514496,"vramReservedPct":37.07654765834442,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":17242.894307728373,"meanTps":17247.578319564556,"stepMs":950.1885070800781,"jitter":0.0015576554007308055,"achievedTflops":143.49095004088227,"nominalPeakTflops":312.0,"mfuNominalPct":45.990689115667394,"configuredPeakTflops":312.0,"mfuConfiguredPct":45.990689115667394,"vramAllocatedGb":52.890898432,"vramAllocatedPct":62.24345000478172,"vramReservedGb":54.184116224,"vramReservedPct":63.765343929218176,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.451170304,"vramAllocatedPct":98.20761039511679,"vramReservedGb":83.537952768,"vramReservedPct":98.30973836267665,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.14 GiB of which 844.75 MiB is free. Process 764857 has 78.30 GiB memory in use. Of the allocated memory 77.72 GiB is allocated by PyTorch, and 82.76 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":16452.89599989703,"meanTps":16458.44447562184,"stepMs":497.90626525878906,"jitter":0.001668259718544926,"achievedTflops":136.91678646379643,"nominalPeakTflops":312.0,"mfuNominalPct":43.88358540506296,"configuredPeakTflops":312.0,"mfuConfiguredPct":43.88358540506296,"vramAllocatedGb":31.420412416,"vramAllocatedPct":36.97639721244883,"vramReservedGb":31.505514496,"vramReservedPct":37.07654765834442,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":17258.61097526465,"meanTps":17257.63340551148,"stepMs":949.3232116699219,"jitter":0.001742568685294182,"achievedTflops":143.62174012264057,"nominalPeakTflops":312.0,"mfuNominalPct":46.03260901366685,"configuredPeakTflops":312.0,"mfuConfiguredPct":46.03260901366685,"vramAllocatedGb":52.890898432,"vramAllocatedPct":62.24345000478172,"vramReservedGb":54.184116224,"vramReservedPct":63.765343929218176,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.451170304,"vramAllocatedPct":98.20761039511679,"vramReservedGb":83.537952768,"vramReservedPct":98.30973836267665,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.14 GiB of which 844.75 MiB is free. Process 852916 has 78.30 GiB memory in use. Of the allocated memory 77.72 GiB is allocated by PyTorch, and 82.76 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":52808.83797736602,"meanTps":52790.7392887647,"stepMs":620.502197265625,"jitter":0.0003722429519152872,"achievedTflops":439.4616238255865,"nominalPeakTflops":2250.0,"mfuNominalPct":19.53162772558162,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.53162772558162,"vramAllocatedGb":95.831739392,"vramAllocatedPct":50.04189778580776,"vramReservedGb":98.484355072,"vramReservedPct":51.42705393100314,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":54684.30600159927,"meanTps":54661.71727159292,"stepMs":1198.4425659179688,"jitter":0.0008714255149772987,"achievedTflops":455.0687883633815,"nominalPeakTflops":2250.0,"mfuNominalPct":20.225279482816955,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.225279482816955,"vramAllocatedGb":181.713421312,"vramAllocatedPct":94.88802471181724,"vramReservedGb":186.883506176,"vramReservedPct":97.58776552785254,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.201229312,"vramAllocatedPct":98.79804580663453,"vramReservedGb":190.293475328,"vramReservedPct":99.36840030334307,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.12 GiB is free. Including non-PyTorch memory, this process has 177.22 GiB memory in use. Of the allocated memory 176.21 GiB is allocated by PyTorch, and 201.65 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":52791.904160343474,"meanTps":52772.502308578034,"stepMs":620.7012329101562,"jitter":0.00034413231278533044,"achievedTflops":439.32070493754964,"nominalPeakTflops":2250.0,"mfuNominalPct":19.525364663891093,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.525364663891093,"vramAllocatedGb":95.831739392,"vramAllocatedPct":50.04189778580776,"vramReservedGb":98.484355072,"vramReservedPct":51.42705393100314,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":54751.82530845208,"meanTps":54741.85005032923,"stepMs":1196.9646606445312,"jitter":0.00013025907970646577,"achievedTflops":455.6306667414256,"nominalPeakTflops":2250.0,"mfuNominalPct":20.250251855174472,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.250251855174472,"vramAllocatedGb":181.713421312,"vramAllocatedPct":94.88802471181724,"vramReservedGb":186.883506176,"vramReservedPct":97.58776552785254,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.201229312,"vramAllocatedPct":98.79804580663453,"vramReservedGb":190.293475328,"vramReservedPct":99.36840030334307,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.12 GiB is free. Including non-PyTorch memory, this process has 177.22 GiB memory in use. Of the allocated memory 176.21 GiB is allocated by PyTorch, and 201.65 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":51964.139606916295,"meanTps":51958.65485158264,"stepMs":630.5887145996094,"jitter":0.00033407258561154166,"achievedTflops":432.43226033760817,"nominalPeakTflops":2250.0,"mfuNominalPct":19.219211570560365,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.219211570560365,"vramAllocatedGb":95.831739392,"vramAllocatedPct":33.341054369509145,"vramReservedGb":98.484355072,"vramReservedPct":34.263932365561416,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":55108.776391595085,"meanTps":55105.8751660207,"stepMs":1189.211669921875,"jitter":0.00012715145788009922,"achievedTflops":458.60112223017467,"nominalPeakTflops":2250.0,"mfuNominalPct":20.382272099118875,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.382272099118875,"vramAllocatedGb":181.713421312,"vramAllocatedPct":63.22035995663746,"vramReservedGb":186.883506176,"vramReservedPct":65.01909680143682,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.273686016,"vramAllocatedPct":99.59817687307918,"vramReservedGb":286.51290624,"vramReservedPct":99.68140456512852,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 24.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 19.62 MiB is free. Including non-PyTorch memory, this process has 267.65 GiB memory in use. Of the allocated memory 266.61 GiB is allocated by PyTorch, and 228.14 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":52029.29478750929,"meanTps":52026.44152400516,"stepMs":629.7990417480469,"jitter":0.0004322669948110831,"achievedTflops":432.97446506243693,"nominalPeakTflops":2250.0,"mfuNominalPct":19.24330955833053,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.24330955833053,"vramAllocatedGb":95.831739392,"vramAllocatedPct":33.341054369509145,"vramReservedGb":98.484355072,"vramReservedPct":34.263932365561416,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":55124.956791715915,"meanTps":55125.758278226924,"stepMs":1188.8626098632812,"jitter":8.304187346704605e-05,"achievedTflops":458.73577137572676,"nominalPeakTflops":2250.0,"mfuNominalPct":20.388256505587854,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.388256505587854,"vramAllocatedGb":181.713421312,"vramAllocatedPct":63.22035995663746,"vramReservedGb":186.883506176,"vramReservedPct":65.01909680143682,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.273686016,"vramAllocatedPct":99.59817687307918,"vramReservedGb":286.51290624,"vramReservedPct":99.68140456512852,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 24.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 19.62 MiB is free. Including non-PyTorch memory, this process has 267.65 GiB memory in use. Of the allocated memory 266.61 GiB is allocated by PyTorch, and 228.14 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":23.663562752,"vramAllocatedPct":93.67609345367659,"vramReservedGb":23.743954944,"vramReservedPct":93.99433912824651,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 987.69 MiB is free. Including non-PyTorch memory, this process has 22.55 GiB memory in use. Of the allocated memory 22.04 GiB is allocated by PyTorch, and 56.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":23.663562752,"vramAllocatedPct":93.67609345367659,"vramReservedGb":23.743954944,"vramReservedPct":93.99433912824651,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 987.69 MiB is free. Including non-PyTorch memory, this process has 22.55 GiB memory in use. Of the allocated memory 22.04 GiB is allocated by PyTorch, and 56.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":20696.213470977356,"meanTps":20695.700716505937,"stepMs":395.8211975097656,"jitter":0.00032709112663738895,"achievedTflops":172.22858762570965,"nominalPeakTflops":209.5,"mfuNominalPct":82.20934970200939,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":64.03815687143579,"vramAllocatedGb":31.420412416,"vramAllocatedPct":93.32188383218555,"vramReservedGb":31.526486016,"vramReservedPct":93.63693342624565,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.91084544,"vramAllocatedPct":97.74862450267253,"vramReservedGb":32.992395264,"vramReservedPct":97.990835938382,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 48.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 35.88 MiB is free. Including non-PyTorch memory, this process has 31.31 GiB memory in use. Of the allocated memory 30.64 GiB is allocated by PyTorch, and 91.83 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":20689.477167969275,"meanTps":20690.038013539903,"stepMs":395.9500732421875,"jitter":0.00017540501778412593,"achievedTflops":172.17252983743222,"nominalPeakTflops":209.5,"mfuNominalPct":82.18259180784355,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":64.01731342442699,"vramAllocatedGb":31.420412416,"vramAllocatedPct":93.32188383218555,"vramReservedGb":31.526486016,"vramReservedPct":93.63693342624565,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.91084544,"vramAllocatedPct":97.74862450267253,"vramReservedGb":32.992395264,"vramReservedPct":97.990835938382,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 48.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 35.88 MiB is free. Including non-PyTorch memory, this process has 31.31 GiB memory in use. Of the allocated memory 30.64 GiB is allocated by PyTorch, and 91.83 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":34531.911717269366,"meanTps":34530.81651938026,"stepMs":237.22984313964844,"jitter":0.00209905772750472,"achievedTflops":287.36572472163135,"nominalPeakTflops":989.5,"mfuNominalPct":29.04150830941196,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.04150830941196,"vramAllocatedGb":31.47048192,"vramAllocatedPct":37.01647816577813,"vramReservedGb":31.568429056,"vramReservedPct":37.131686379950416,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":37275.29116954039,"meanTps":37297.99919364846,"stepMs":439.5404968261719,"jitter":0.0007826366049583336,"achievedTflops":310.19542586713845,"nominalPeakTflops":989.5,"mfuNominalPct":31.34870397848797,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.34870397848797,"vramAllocatedGb":52.940967936,"vramAllocatedPct":62.27067601505939,"vramReservedGb":54.247030784,"vramReservedPct":63.80690437721235,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.501239808,"vramAllocatedPct":98.2165391691668,"vramReservedGb":83.579895808,"vramReservedPct":98.30905659911937,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.18 GiB of which 620.19 MiB is free. Including non-PyTorch memory, this process has 78.56 GiB memory in use. Of the allocated memory 77.77 GiB is allocated by PyTorch, and 75.01 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":34825.43749109897,"meanTps":34811.08643846779,"stepMs":235.23035430908203,"jitter":0.0005469808378037535,"achievedTflops":289.8083710312719,"nominalPeakTflops":989.5,"mfuNominalPct":29.288364934944106,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.288364934944106,"vramAllocatedGb":31.47048192,"vramAllocatedPct":37.01647816577813,"vramReservedGb":31.568429056,"vramReservedPct":37.131686379950416,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":37430.28019954648,"meanTps":37431.86299739699,"stepMs":437.72047424316406,"jitter":0.00014269659517634963,"achievedTflops":311.4852048778189,"nominalPeakTflops":989.5,"mfuNominalPct":31.479050518223236,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.479050518223236,"vramAllocatedGb":52.940967936,"vramAllocatedPct":62.27067601505939,"vramReservedGb":54.247030784,"vramReservedPct":63.80690437721235,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.501239808,"vramAllocatedPct":98.2165391691668,"vramReservedGb":83.579895808,"vramReservedPct":98.30905659911937,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.18 GiB of which 620.19 MiB is free. Including non-PyTorch memory, this process has 78.56 GiB memory in use. Of the allocated memory 77.77 GiB is allocated by PyTorch, and 75.01 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":40041.860293920494,"meanTps":40043.13540924569,"stepMs":409.1717987060547,"jitter":0.0014942945536353008,"achievedTflops":333.218105524413,"nominalPeakTflops":989.5,"mfuNominalPct":33.6754022763429,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.6754022763429,"vramAllocatedGb":52.940967936,"vramAllocatedPct":35.26558764042344,"vramReservedGb":54.194601984,"vramReservedPct":36.10067137826156,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":41607.406380104054,"meanTps":41550.3626173709,"stepMs":787.5520935058594,"jitter":0.0062351165153811785,"achievedTflops":346.24617907344464,"nominalPeakTflops":989.5,"mfuNominalPct":34.992034267149535,"configuredPeakTflops":989.5,"mfuConfiguredPct":34.992034267149535,"vramAllocatedGb":95.881808896,"vramAllocatedPct":63.86978679407385,"vramReservedGb":98.38788608,"vramReservedPct":65.53916096707457,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.296026624,"vramAllocatedPct":98.7844901127884,"vramReservedGb":149.011038208,"vramReservedPct":99.26078105839319,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 749.44 MiB is free. Including non-PyTorch memory, this process has 139.07 GiB memory in use. Of the allocated memory 138.11 GiB is allocated by PyTorch, and 241.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":40130.110526734185,"meanTps":40165.22154683954,"stepMs":408.27198791503906,"jitter":0.0017064866486899844,"achievedTflops":333.95250135853286,"nominalPeakTflops":989.5,"mfuNominalPct":33.74962115801241,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.74962115801241,"vramAllocatedGb":52.940967936,"vramAllocatedPct":35.26558764042344,"vramReservedGb":54.194601984,"vramReservedPct":36.10067137826156,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":41656.645679316534,"meanTps":41570.654300793074,"stepMs":786.6211853027344,"jitter":0.003677970335431758,"achievedTflops":346.6559359099276,"nominalPeakTflops":989.5,"mfuNominalPct":35.03344476098309,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.03344476098309,"vramAllocatedGb":95.881808896,"vramAllocatedPct":63.86978679407385,"vramReservedGb":98.38788608,"vramReservedPct":65.53916096707457,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.296026624,"vramAllocatedPct":98.7844901127884,"vramReservedGb":149.011038208,"vramReservedPct":99.26078105839319,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 749.44 MiB is free. Including non-PyTorch memory, this process has 139.07 GiB memory in use. Of the allocated memory 138.11 GiB is allocated by PyTorch, and 241.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":12831.911636327792,"meanTps":12839.938928229418,"stepMs":638.4083862304688,"jitter":0.0023260232647585974,"achievedTflops":106.7838820256563,"nominalPeakTflops":364.2,"mfuNominalPct":29.320121368933634,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":23.65439743290781,"vramAllocatedGb":31.420412416,"vramAllocatedPct":61.7577883978751,"vramReservedGb":31.526486016,"vramReservedPct":61.96627932589567,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":49.745552896,"vramAllocatedPct":97.77641645219298,"vramReservedGb":49.968840704,"vramReservedPct":98.21529551640498,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.38 GiB of which 400.50 MiB is free. Process 2319972 has 46.98 GiB memory in use. Of the allocated memory 46.33 GiB is allocated by PyTorch, and 152.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":9233.503587106943,"meanTps":9226.64349980555,"stepMs":887.203857421875,"jitter":0.0008646661022727929,"achievedTflops":76.83885189309721,"nominalPeakTflops":154.8,"mfuNominalPct":49.637501222931014,"configuredPeakTflops":180.6,"mfuConfiguredPct":42.54642961965516,"vramAllocatedGb":31.420412416,"vramAllocatedPct":61.567689030724935,"vramReservedGb":31.505514496,"vramReservedPct":61.734444906743924,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.745552896,"vramAllocatedPct":97.47544656042768,"vramReservedGb":49.989812224,"vramReservedPct":97.95406797856985,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 738.25 MiB is free. Process 1415479 has 46.80 GiB memory in use. Of the allocated memory 46.33 GiB is allocated by PyTorch, and 152.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":9229.043015171554,"meanTps":9229.804508750552,"stepMs":887.6326599121094,"jitter":0.0016612123373531552,"achievedTflops":76.80173215592828,"nominalPeakTflops":154.8,"mfuNominalPct":49.613522064553145,"configuredPeakTflops":180.6,"mfuConfiguredPct":42.525876055331274,"vramAllocatedGb":31.420412416,"vramAllocatedPct":61.567689030724935,"vramReservedGb":31.505514496,"vramReservedPct":61.734444906743924,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":49.745552896,"vramAllocatedPct":97.47544656042768,"vramReservedGb":49.989812224,"vramReservedPct":97.95406797856985,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 738.25 MiB is free. Process 1449495 has 46.80 GiB memory in use. Of the allocated memory 46.33 GiB is allocated by PyTorch, and 152.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":26730.632734267183,"meanTps":26729.256644439705,"stepMs":306.4648742675781,"jitter":0.0006392654786914594,"achievedTflops":222.44547915107054,"nominalPeakTflops":468.0,"mfuNominalPct":47.531085288690285,"configuredPeakTflops":468.0,"mfuConfiguredPct":47.531085288690285,"vramAllocatedGb":31.420412416,"vramAllocatedPct":30.812194950800098,"vramReservedGb":31.526486016,"vramReservedPct":30.916215241783576,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":26417.197134585076,"meanTps":26415.479323549564,"stepMs":620.2020568847656,"jitter":0.0012697402486401836,"achievedTflops":219.83714837014932,"nominalPeakTflops":468.0,"mfuNominalPct":46.97374965174131,"configuredPeakTflops":468.0,"mfuConfiguredPct":46.97374965174131,"vramAllocatedGb":52.890898432,"vramAllocatedPct":51.867068198629944,"vramReservedGb":54.184116224,"vramReservedPct":53.135252657617386,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":26646.00279368871,"meanTps":26638.278088196057,"stepMs":307.438232421875,"jitter":0.0012144381612477734,"achievedTflops":221.7412104616739,"nominalPeakTflops":468.0,"mfuNominalPct":47.38060052599869,"configuredPeakTflops":468.0,"mfuConfiguredPct":47.38060052599869,"vramAllocatedGb":31.43803392,"vramAllocatedPct":30.829475348313206,"vramReservedGb":33.288093696,"vramReservedPct":32.643722778742145,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":26399.89925303759,"meanTps":26398.634367778246,"stepMs":620.6084289550781,"jitter":0.00021536091217503532,"achievedTflops":219.69319983038295,"nominalPeakTflops":468.0,"mfuNominalPct":46.94299141674849,"configuredPeakTflops":468.0,"mfuConfiguredPct":46.94299141674849,"vramAllocatedGb":52.908839936,"vramAllocatedPct":51.88466240177532,"vramReservedGb":56.488886272,"vramReservedPct":55.39540835180485,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":13660.175649505427,"meanTps":13660.1484000461,"stepMs":1199.39892578125,"jitter":0.0007159139872714563,"achievedTflops":169.36715434685078,"nominalPeakTflops":312.0,"mfuNominalPct":54.284344341939345,"configuredPeakTflops":312.0,"mfuConfiguredPct":54.284344341939345,"vramAllocatedGb":52.919398912,"vramAllocatedPct":62.27699014598906,"vramReservedGb":53.108277248,"vramReservedPct":62.4992673167752,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.479481856,"vramAllocatedPct":98.24092820070739,"vramReservedGb":83.630227456,"vramReservedPct":98.41832972904604,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.14 GiB of which 816.75 MiB is free. Process 925444 has 78.33 GiB memory in use. Of the allocated memory 77.75 GiB is allocated by PyTorch, and 83.76 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":13642.294349121086,"meanTps":13642.883212874302,"stepMs":1200.9710083007812,"jitter":0.0012302815882473028,"achievedTflops":169.1454511242991,"nominalPeakTflops":312.0,"mfuNominalPct":54.21328561676252,"configuredPeakTflops":312.0,"mfuConfiguredPct":54.21328561676252,"vramAllocatedGb":52.919398912,"vramAllocatedPct":62.27699014598906,"vramReservedGb":53.108277248,"vramReservedPct":62.4992673167752,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.479481856,"vramAllocatedPct":98.24092820070739,"vramReservedGb":83.630227456,"vramReservedPct":98.41832972904604,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.14 GiB of which 816.75 MiB is free. Process 937161 has 78.33 GiB memory in use. Of the allocated memory 77.75 GiB is allocated by PyTorch, and 83.76 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":37446.07763866427,"meanTps":37439.80353784656,"stepMs":875.0716247558594,"jitter":0.0005049486078968075,"achievedTflops":464.27921381387426,"nominalPeakTflops":2250.0,"mfuNominalPct":20.634631725061077,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.634631725061077,"vramAllocatedGb":95.860690944,"vramAllocatedPct":50.05701585227631,"vramReservedGb":98.509520896,"vramReservedPct":51.44019514705842,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":38613.83924782581,"meanTps":38610.260389398994,"stepMs":1697.21533203125,"jitter":0.0005854758714081573,"achievedTflops":478.75783149595,"nominalPeakTflops":2250.0,"mfuNominalPct":21.27812584426444,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.27812584426444,"vramAllocatedGb":181.743012864,"vramAllocatedPct":94.90347697669213,"vramReservedGb":186.910769152,"vramReservedPct":97.60200184524575,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.229540864,"vramAllocatedPct":98.81282967469673,"vramReservedGb":190.320738304,"vramReservedPct":99.38263662073629,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.09 GiB is free. Including non-PyTorch memory, this process has 177.25 GiB memory in use. Of the allocated memory 176.23 GiB is allocated by PyTorch, and 200.65 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":37532.442836212416,"meanTps":37526.60141670351,"stepMs":873.0580139160156,"jitter":0.00021719439751275544,"achievedTflops":465.3500219878425,"nominalPeakTflops":2250.0,"mfuNominalPct":20.68222319945967,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.68222319945967,"vramAllocatedGb":95.860690944,"vramAllocatedPct":50.05701585227631,"vramReservedGb":98.509520896,"vramReservedPct":51.44019514705842,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":38639.49175051815,"meanTps":38636.967787548834,"stepMs":1696.0885620117188,"jitter":0.00020695470805325112,"achievedTflops":479.0758868046336,"nominalPeakTflops":2250.0,"mfuNominalPct":21.292261635761495,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.292261635761495,"vramAllocatedGb":181.743012864,"vramAllocatedPct":94.90347697669213,"vramReservedGb":186.910769152,"vramReservedPct":97.60200184524575,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.229540864,"vramAllocatedPct":98.81282967469673,"vramReservedGb":190.320738304,"vramReservedPct":99.38263662073629,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.09 GiB is free. Including non-PyTorch memory, this process has 177.25 GiB memory in use. Of the allocated memory 176.23 GiB is allocated by PyTorch, and 200.65 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":37288.67617888505,"meanTps":37267.87426674834,"stepMs":878.7654418945312,"jitter":0.0005949885721274929,"achievedTflops":462.32765491618085,"nominalPeakTflops":2250.0,"mfuNominalPct":20.54789577405248,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.54789577405248,"vramAllocatedGb":95.860690944,"vramAllocatedPct":33.35112697463389,"vramReservedGb":98.509520896,"vramReservedPct":34.27268786881702,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":38980.24166808461,"meanTps":38978.90769265882,"stepMs":1681.2620239257812,"jitter":0.0001351497870981914,"achievedTflops":483.3007112404925,"nominalPeakTflops":2250.0,"mfuNominalPct":21.480031610688552,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.480031610688552,"vramAllocatedGb":181.743012864,"vramAllocatedPct":63.2306552257244,"vramReservedGb":186.910769152,"vramReservedPct":65.02858192996372,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.4966912,"vramAllocatedPct":99.3278507100622,"vramReservedGb":286.372397056,"vramReservedPct":99.63251967195136,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 768.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 753.62 MiB is free. Including non-PyTorch memory, this process has 266.94 GiB memory in use. Of the allocated memory 265.89 GiB is allocated by PyTorch, and 235.14 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":37248.19203097352,"meanTps":37211.03802674748,"stepMs":879.7205505371094,"jitter":0.0011460998964050078,"achievedTflops":461.82570786192167,"nominalPeakTflops":2250.0,"mfuNominalPct":20.52558701608541,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.52558701608541,"vramAllocatedGb":95.860690944,"vramAllocatedPct":33.35112697463389,"vramReservedGb":98.509520896,"vramReservedPct":34.27268786881702,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":38970.52230914706,"meanTps":38965.127116008494,"stepMs":1681.6813354492188,"jitter":0.00010261504630386597,"achievedTflops":483.18020472523494,"nominalPeakTflops":2250.0,"mfuNominalPct":21.474675765565998,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.474675765565998,"vramAllocatedGb":181.743012864,"vramAllocatedPct":63.2306552257244,"vramReservedGb":186.910769152,"vramReservedPct":65.02858192996372,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.4966912,"vramAllocatedPct":99.3278507100622,"vramReservedGb":286.372397056,"vramReservedPct":99.63251967195136,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 768.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 753.62 MiB is free. Including non-PyTorch memory, this process has 266.94 GiB memory in use. Of the allocated memory 265.89 GiB is allocated by PyTorch, and 235.14 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.856125952,"vramAllocatedPct":97.5861019477524,"vramReservedGb":32.99868672,"vramReservedPct":98.00952221525813,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 48.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 29.88 MiB is free. Including non-PyTorch memory, this process has 31.32 GiB memory in use. Of the allocated memory 30.59 GiB is allocated by PyTorch, and 150.08 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.856125952,"vramAllocatedPct":97.5861019477524,"vramReservedGb":32.99868672,"vramReservedPct":98.00952221525813,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 48.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 29.88 MiB is free. Including non-PyTorch memory, this process has 31.32 GiB memory in use. Of the allocated memory 30.59 GiB is allocated by PyTorch, and 150.08 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":28944.390225808667,"meanTps":28940.71575577715,"stepMs":566.0509643554688,"jitter":0.0002720072472183252,"achievedTflops":358.8701296844234,"nominalPeakTflops":989.5,"mfuNominalPct":36.26782513233182,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.26782513233182,"vramAllocatedGb":52.969468416,"vramAllocatedPct":62.30419909228191,"vramReservedGb":53.171191808,"vramReservedPct":62.54147189777871,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.52955136,"vramAllocatedPct":98.24984002388874,"vramReservedGb":83.693142016,"vramReservedPct":98.44226001800713,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.18 GiB of which 572.19 MiB is free. Including non-PyTorch memory, this process has 78.61 GiB memory in use. Of the allocated memory 77.79 GiB is allocated by PyTorch, and 96.01 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":28959.266145706064,"meanTps":28958.135707441543,"stepMs":565.7601928710938,"jitter":0.00020305064559090298,"achievedTflops":359.0545703743501,"nominalPeakTflops":989.5,"mfuNominalPct":36.286464919085404,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.286464919085404,"vramAllocatedGb":52.969468416,"vramAllocatedPct":62.30419909228191,"vramReservedGb":53.171191808,"vramReservedPct":62.54147189777871,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.52955136,"vramAllocatedPct":98.24984002388874,"vramReservedGb":83.693142016,"vramReservedPct":98.44226001800713,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 79.18 GiB of which 572.19 MiB is free. Including non-PyTorch memory, this process has 78.61 GiB memory in use. Of the allocated memory 77.79 GiB is allocated by PyTorch, and 96.01 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":30579.316758738045,"meanTps":30568.05387976025,"stepMs":535.7869873046875,"jitter":0.0043152185824781785,"achievedTflops":379.14094182866086,"nominalPeakTflops":989.5,"mfuNominalPct":38.31641655671156,"configuredPeakTflops":989.5,"mfuConfiguredPct":38.31641655671156,"vramAllocatedGb":52.969468416,"vramAllocatedPct":35.284572676293,"vramReservedGb":53.171191808,"vramReservedPct":35.41894675816282,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":31332.05885657176,"meanTps":31328.98956223619,"stepMs":1045.8297729492188,"jitter":0.0005650966581314424,"achievedTflops":388.4738955430438,"nominalPeakTflops":989.5,"mfuNominalPct":39.259615517235346,"configuredPeakTflops":989.5,"mfuConfiguredPct":39.259615517235346,"vramAllocatedGb":95.910760448,"vramAllocatedPct":63.88907230271088,"vramReservedGb":98.415149056,"vramReservedPct":65.55732166392146,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.324338176,"vramAllocatedPct":98.80334929797556,"vramReservedGb":149.040398336,"vramReservedPct":99.28033873192061,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 721.44 MiB is free. Including non-PyTorch memory, this process has 139.10 GiB memory in use. Of the allocated memory 138.14 GiB is allocated by PyTorch, and 242.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":30466.351215877003,"meanTps":30505.31066060809,"stepMs":537.7736206054688,"jitter":0.0030039412848859755,"achievedTflops":377.74032641751757,"nominalPeakTflops":989.5,"mfuNominalPct":38.17486876377136,"configuredPeakTflops":989.5,"mfuConfiguredPct":38.17486876377136,"vramAllocatedGb":52.969468416,"vramAllocatedPct":35.284572676293,"vramReservedGb":53.171191808,"vramReservedPct":35.41894675816282,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":31320.806315559432,"meanTps":31327.510072582343,"stepMs":1046.2055053710938,"jitter":0.0007617545409024695,"achievedTflops":388.3343797052297,"nominalPeakTflops":989.5,"mfuNominalPct":39.24551588734004,"configuredPeakTflops":989.5,"mfuConfiguredPct":39.24551588734004,"vramAllocatedGb":95.910760448,"vramAllocatedPct":63.88907230271088,"vramReservedGb":98.415149056,"vramReservedPct":65.55732166392146,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.324338176,"vramAllocatedPct":98.80334929797556,"vramReservedGb":149.040398336,"vramReservedPct":99.28033873192061,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 721.44 MiB is free. Including non-PyTorch memory, this process has 139.10 GiB memory in use. Of the allocated memory 138.14 GiB is allocated by PyTorch, and 242.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":48.725349376,"vramAllocatedPct":95.4763775400788,"vramReservedGb":48.913973248,"vramReservedPct":95.8459823607066,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.72 GiB is free. Process 1472317 has 45.80 GiB memory in use. Of the allocated memory 45.38 GiB is allocated by PyTorch, and 99.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":48.725349376,"vramAllocatedPct":95.4763775400788,"vramReservedGb":48.913973248,"vramReservedPct":95.8459823607066,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 47.53 GiB of which 1.72 GiB is free. Process 1479887 has 45.80 GiB memory in use. Of the allocated memory 45.38 GiB is allocated by PyTorch, and 99.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":22407.481574388457,"meanTps":22406.768394315848,"stepMs":731.1843566894531,"jitter":0.00022790239458344804,"achievedTflops":277.82156596727714,"nominalPeakTflops":468.0,"mfuNominalPct":59.36358247164041,"configuredPeakTflops":468.0,"mfuConfiguredPct":59.36358247164041,"vramAllocatedGb":52.935840768,"vramAllocatedPct":51.911140567892396,"vramReservedGb":55.905878016,"vramReservedPct":54.82368561933523,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":22617.5826034109,"meanTps":22617.782798026426,"stepMs":1448.7843627929688,"jitter":0.0001522144335463085,"achievedTflops":280.42652613205837,"nominalPeakTflops":468.0,"mfuNominalPct":59.92019789146546,"configuredPeakTflops":468.0,"mfuConfiguredPct":59.92019789146546,"vramAllocatedGb":95.897375744,"vramAllocatedPct":94.041051962437,"vramReservedGb":98.511618048,"vramReservedPct":96.60468933463325,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":22401.49679648333,"meanTps":22403.01143499722,"stepMs":731.3796997070312,"jitter":0.0002137367545903859,"achievedTflops":277.74736305589465,"nominalPeakTflops":468.0,"mfuNominalPct":59.347727148695434,"configuredPeakTflops":468.0,"mfuConfiguredPct":59.347727148695434,"vramAllocatedGb":52.935840768,"vramAllocatedPct":51.911140567892396,"vramReservedGb":55.905878016,"vramReservedPct":54.82368561933523,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":22621.579543725536,"meanTps":22621.847867731074,"stepMs":1448.5283813476562,"jitter":0.00011500873752653986,"achievedTflops":280.47608262566087,"nominalPeakTflops":468.0,"mfuNominalPct":59.93078688582497,"configuredPeakTflops":468.0,"mfuConfiguredPct":59.93078688582497,"vramAllocatedGb":95.897375744,"vramAllocatedPct":94.041051962437,"vramReservedGb":98.511618048,"vramReservedPct":96.60468933463325,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":81.86906624,"vramAllocatedPct":96.34574723662737,"vramReservedGb":82.11398656,"vramReservedPct":96.63397614074921,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 79.14 GiB of which 2.33 GiB is free. Process 949988 has 76.80 GiB memory in use. Of the allocated memory 76.25 GiB is allocated by PyTorch, and 53.57 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":81.86906624,"vramAllocatedPct":96.34574723662737,"vramReservedGb":82.11398656,"vramReservedPct":96.63397614074921,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 79.14 GiB of which 2.33 GiB is free. Process 970883 has 76.80 GiB memory in use. Of the allocated memory 76.25 GiB is allocated by PyTorch, and 53.57 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":23875.75255055016,"meanTps":23803.926635187472,"stepMs":1372.4384155273438,"jitter":0.004878608376907423,"achievedTflops":490.7024489425117,"nominalPeakTflops":2250.0,"mfuNominalPct":21.808997730778298,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.808997730778298,"vramAllocatedGb":95.917371904,"vramAllocatedPct":50.08661379993662,"vramReservedGb":96.39559168,"vramReservedPct":50.33633299841484,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":24407.39360875342,"meanTps":24397.64730133495,"stepMs":2685.0880126953125,"jitter":0.0022401196186202905,"achievedTflops":501.62891371744985,"nominalPeakTflops":2250.0,"mfuNominalPct":22.294618387442213,"configuredPeakTflops":2250.0,"mfuConfiguredPct":22.294618387442213,"vramAllocatedGb":181.800275968,"vramAllocatedPct":94.93337891122287,"vramReservedGb":187.168718848,"vramReservedPct":97.73669930981238,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.286163968,"vramAllocatedPct":98.8423974108211,"vramReservedGb":190.576590848,"vramReservedPct":99.51623898396498,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 893.81 MiB is free. Including non-PyTorch memory, this process has 177.46 GiB memory in use. Of the allocated memory 176.29 GiB is allocated by PyTorch, and 368.65 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":23959.976771532165,"meanTps":23943.75139036486,"stepMs":1367.614013671875,"jitter":0.001034452662713971,"achievedTflops":492.43345329132296,"nominalPeakTflops":2250.0,"mfuNominalPct":21.88593125739213,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.88593125739213,"vramAllocatedGb":95.917371904,"vramAllocatedPct":50.08661379993662,"vramReservedGb":96.39559168,"vramReservedPct":50.33633299841484,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":24444.37092675138,"meanTps":24412.585721283984,"stepMs":2681.0262451171875,"jitter":0.0009802334356737866,"achievedTflops":502.3888839197926,"nominalPeakTflops":2250.0,"mfuNominalPct":22.32839484087967,"configuredPeakTflops":2250.0,"mfuConfiguredPct":22.32839484087967,"vramAllocatedGb":181.800275968,"vramAllocatedPct":94.93337891122287,"vramReservedGb":187.168718848,"vramReservedPct":97.73669930981238,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.286163968,"vramAllocatedPct":98.8423974108211,"vramReservedGb":190.576590848,"vramReservedPct":99.51623898396498,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 2.00 GiB. GPU 0 has a total capacity of 178.35 GiB of which 893.75 MiB is free. Including non-PyTorch memory, this process has 177.46 GiB memory in use. Of the allocated memory 176.29 GiB is allocated by PyTorch, and 368.65 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":24029.068826490784,"meanTps":24017.083052356484,"stepMs":1363.681640625,"jitter":0.00019920569531871012,"achievedTflops":493.85345630479384,"nominalPeakTflops":2250.0,"mfuNominalPct":21.949042502435283,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.949042502435283,"vramAllocatedGb":95.917371904,"vramAllocatedPct":33.3708469857812,"vramReservedGb":96.39559168,"vramReservedPct":33.53722559534572,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":24730.277824647626,"meanTps":24726.793028545704,"stepMs":2650.0308837890625,"jitter":9.731907697678014e-05,"achievedTflops":508.26493807432473,"nominalPeakTflops":2250.0,"mfuNominalPct":22.589552803303324,"configuredPeakTflops":2250.0,"mfuConfiguredPct":22.589552803303324,"vramAllocatedGb":181.800275968,"vramAllocatedPct":63.25057777201171,"vramReservedGb":187.168718848,"vramReservedPct":65.11832583833373,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.553314304,"vramAllocatedPct":99.34755059238732,"vramReservedGb":286.4185344,"vramReservedPct":99.64857142791998,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 768.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 569.62 MiB is free. Including non-PyTorch memory, this process has 267.12 GiB memory in use. Of the allocated memory 265.94 GiB is allocated by PyTorch, and 365.14 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":24037.03542688645,"meanTps":24032.733344610395,"stepMs":1363.2296752929688,"jitter":0.00019130618077197134,"achievedTflops":494.01718853964684,"nominalPeakTflops":2250.0,"mfuNominalPct":21.95631949065097,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.95631949065097,"vramAllocatedGb":95.917371904,"vramAllocatedPct":33.3708469857812,"vramReservedGb":96.39559168,"vramReservedPct":33.53722559534572,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":24738.65811151468,"meanTps":24737.48277411421,"stepMs":2649.1331787109375,"jitter":8.40985122869873e-05,"achievedTflops":508.43717253184803,"nominalPeakTflops":2250.0,"mfuNominalPct":22.597207668082135,"configuredPeakTflops":2250.0,"mfuConfiguredPct":22.597207668082135,"vramAllocatedGb":181.800275968,"vramAllocatedPct":63.25057777201171,"vramReservedGb":187.168718848,"vramReservedPct":65.11832583833373,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.553314304,"vramAllocatedPct":99.34755059238732,"vramReservedGb":286.4185344,"vramReservedPct":99.64857142791998,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 768.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 569.62 MiB is free. Including non-PyTorch memory, this process has 267.12 GiB memory in use. Of the allocated memory 265.94 GiB is allocated by PyTorch, and 365.14 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.739265024,"vramAllocatedPct":97.23901287065982,"vramReservedGb":32.864468992,"vramReservedPct":97.61088164190086,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 197.88 MiB is free. Including non-PyTorch memory, this process has 31.15 GiB memory in use. Of the allocated memory 30.49 GiB is allocated by PyTorch, and 79.40 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_extension_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.739265024,"vramAllocatedPct":97.23901287065982,"vramReservedGb":32.864468992,"vramReservedPct":97.61088164190086,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 197.88 MiB is free. Including non-PyTorch memory, this process has 31.15 GiB memory in use. Of the allocated memory 30.49 GiB is allocated by PyTorch, and 79.40 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_32k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":81.919135744,"vramAllocatedPct":96.35562325787195,"vramReservedGb":82.23981568,"vramReservedPct":96.73281614228098,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 79.18 GiB of which 2.07 GiB is free. Including non-PyTorch memory, this process has 77.10 GiB memory in use. Of the allocated memory 76.29 GiB is allocated by PyTorch, and 85.82 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_32k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":81.919135744,"vramAllocatedPct":96.35562325787195,"vramReservedGb":82.23981568,"vramReservedPct":96.73281614228098,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 79.18 GiB of which 2.07 GiB is free. Including non-PyTorch memory, this process has 77.10 GiB memory in use. Of the allocated memory 76.29 GiB is allocated by PyTorch, and 85.82 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":20879.28088311737,"meanTps":20879.834038756562,"stepMs":1569.4027099609375,"jitter":0.0007080566811601901,"achievedTflops":429.1179614050816,"nominalPeakTflops":989.5,"mfuNominalPct":43.36715122840643,"configuredPeakTflops":989.5,"mfuConfiguredPct":43.36715122840643,"vramAllocatedGb":95.967441408,"vramAllocatedPct":63.92682921272506,"vramReservedGb":96.4165632,"vramReservedPct":64.22600288737618,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.38096128,"vramAllocatedPct":98.84106766834988,"vramReservedGb":149.019426816,"vramReservedPct":99.2663689651153,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 681.44 MiB is free. Including non-PyTorch memory, this process has 139.14 GiB memory in use. Of the allocated memory 138.19 GiB is allocated by PyTorch, and 228.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":20915.013515827784,"meanTps":20909.939574517433,"stepMs":1566.721435546875,"jitter":0.0013343256991150936,"achievedTflops":429.8523504192514,"nominalPeakTflops":989.5,"mfuNominalPct":43.441369420844005,"configuredPeakTflops":989.5,"mfuConfiguredPct":43.441369420844005,"vramAllocatedGb":95.967441408,"vramAllocatedPct":63.92682921272506,"vramReservedGb":96.4165632,"vramReservedPct":64.22600288737618,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.38096128,"vramAllocatedPct":98.84106766834988,"vramReservedGb":149.019426816,"vramReservedPct":99.2663689651153,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 681.44 MiB is free. Including non-PyTorch memory, this process has 139.14 GiB memory in use. Of the allocated memory 138.19 GiB is allocated by PyTorch, and 228.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":17411.657063034294,"meanTps":17411.81337071527,"stepMs":1881.9575805664062,"jitter":6.693194747757775e-05,"achievedTflops":357.85019730325666,"nominalPeakTflops":468.0,"mfuNominalPct":76.46371737249073,"configuredPeakTflops":468.0,"mfuConfiguredPct":76.46371737249073,"vramAllocatedGb":95.973659648,"vramAllocatedPct":94.11585920845708,"vramReservedGb":99.1428608,"vramReservedPct":97.22371286871007,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.477937152,"vramAllocatedPct":98.53294561484294,"vramReservedGb":101.235818496,"vramReservedPct":99.27615634714418,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 35.75 MiB is free. Including non-PyTorch memory, this process has 94.93 GiB memory in use. Of the allocated memory 93.58 GiB is allocated by PyTorch, and 722.77 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":17405.32918522209,"meanTps":17405.05438523178,"stepMs":1882.6417846679688,"jitter":0.00024398210639106508,"achievedTflops":357.7201446428229,"nominalPeakTflops":468.0,"mfuNominalPct":76.43592834248354,"configuredPeakTflops":468.0,"mfuConfiguredPct":76.43592834248354,"vramAllocatedGb":95.973659648,"vramAllocatedPct":94.11585920845708,"vramReservedGb":99.1428608,"vramReservedPct":97.22371286871007,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"700m","modelLabel":"700M","parameters":707480064,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.477937152,"vramAllocatedPct":98.53294561484294,"vramReservedGb":101.235818496,"vramReservedPct":99.27615634714418,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 512.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 35.75 MiB is free. Including non-PyTorch memory, this process has 94.93 GiB memory in use. Of the allocated memory 93.58 GiB is allocated by PyTorch, and 722.77 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":16904.73315665686,"meanTps":16900.970025634233,"stepMs":484.59800720214844,"jitter":0.0013723051083934232,"achievedTflops":106.14638716798245,"nominalPeakTflops":312.0,"mfuNominalPct":34.02127793845591,"configuredPeakTflops":312.0,"mfuConfiguredPct":34.02127793845591,"vramAllocatedGb":39.403357696,"vramAllocatedPct":46.370944670655035,"vramReservedGb":40.072380416,"vramReservedPct":47.15826883150137,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":18081.137209891313,"meanTps":18083.66245175937,"stepMs":906.1376953125,"jitter":0.0012987340160688016,"achievedTflops":113.53313731324785,"nominalPeakTflops":312.0,"mfuNominalPct":36.38882606193841,"configuredPeakTflops":312.0,"mfuConfiguredPct":36.38882606193841,"vramAllocatedGb":64.855991808,"vramAllocatedPct":76.32429781471444,"vramReservedGb":66.171437056,"vramReservedPct":77.87234961484,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.316592128,"vramAllocatedPct":99.22606237139482,"vramReservedGb":84.393590784,"vramReservedPct":99.3166764871927,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 28.75 MiB is free. Process 764857 has 79.10 GiB memory in use. Of the allocated memory 78.53 GiB is allocated by PyTorch, and 73.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":16885.809697527908,"meanTps":16889.163072939984,"stepMs":485.1410827636719,"jitter":0.0016122715881000456,"achievedTflops":106.02756501322584,"nominalPeakTflops":312.0,"mfuNominalPct":33.98319391449546,"configuredPeakTflops":312.0,"mfuConfiguredPct":33.98319391449546,"vramAllocatedGb":39.403357696,"vramAllocatedPct":46.370944670655035,"vramReservedGb":40.072380416,"vramReservedPct":47.15826883150137,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":18057.868159387803,"meanTps":18065.96353384372,"stepMs":907.3053283691406,"jitter":0.0018234676140368733,"achievedTflops":113.38702878725766,"nominalPeakTflops":312.0,"mfuNominalPct":36.34199640617233,"configuredPeakTflops":312.0,"mfuConfiguredPct":36.34199640617233,"vramAllocatedGb":64.855991808,"vramAllocatedPct":76.32429781471444,"vramReservedGb":66.171437056,"vramReservedPct":77.87234961484,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.316592128,"vramAllocatedPct":99.22606237139482,"vramReservedGb":84.393590784,"vramReservedPct":99.3166764871927,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 28.75 MiB is free. Process 852916 has 79.10 GiB memory in use. Of the allocated memory 78.53 GiB is allocated by PyTorch, and 73.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":62546.798263996396,"meanTps":62569.39422587111,"stepMs":261.94786071777344,"jitter":0.00808386360703943,"achievedTflops":392.73714663951716,"nominalPeakTflops":2250.0,"mfuNominalPct":17.454984295089652,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.454984295089652,"vramAllocatedGb":64.855991808,"vramAllocatedPct":33.866826726136374,"vramReservedGb":66.13368832,"vramReservedPct":34.53402069193978,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":68578.49471245402,"meanTps":68553.59217955862,"stepMs":477.8174285888672,"jitter":0.0004735060208864062,"achievedTflops":430.6107279947844,"nominalPeakTflops":2250.0,"mfuNominalPct":19.138254577545972,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.138254577545972,"vramAllocatedGb":115.76125952,"vramAllocatedPct":60.44879445170329,"vramReservedGb":118.252109824,"vramReservedPct":61.74947914242614,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.334312448,"vramAllocatedPct":99.38972483525569,"vramReservedGb":190.582882304,"vramReservedPct":99.5195242879788,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 23.81 MiB is free. Including non-PyTorch memory, this process has 178.31 GiB memory in use. Of the allocated memory 177.26 GiB is allocated by PyTorch, and 237.05 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":62599.41717239705,"meanTps":62617.22881543007,"stepMs":261.72767639160156,"jitter":0.001670547628826792,"achievedTflops":393.067545645032,"nominalPeakTflops":2250.0,"mfuNominalPct":17.469668695334757,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.469668695334757,"vramAllocatedGb":64.855991808,"vramAllocatedPct":33.866826726136374,"vramReservedGb":66.13368832,"vramReservedPct":34.53402069193978,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":68552.97964642546,"meanTps":68433.96497838169,"stepMs":477.9952697753906,"jitter":0.001096784915790762,"achievedTflops":430.45051652902595,"nominalPeakTflops":2250.0,"mfuNominalPct":19.131134067956708,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.131134067956708,"vramAllocatedGb":115.76125952,"vramAllocatedPct":60.44879445170329,"vramReservedGb":118.252109824,"vramReservedPct":61.74947914242614,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.334312448,"vramAllocatedPct":99.38972483525569,"vramReservedGb":190.582882304,"vramReservedPct":99.5195242879788,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 23.81 MiB is free. Including non-PyTorch memory, this process has 178.31 GiB memory in use. Of the allocated memory 177.26 GiB is allocated by PyTorch, and 237.05 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":66073.98434974613,"meanTps":66068.03417699128,"stepMs":495.9289245605469,"jitter":0.00023159639223144942,"achievedTflops":414.88467516906854,"nominalPeakTflops":2250.0,"mfuNominalPct":18.439318896403048,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.439318896403048,"vramAllocatedGb":115.76125952,"vramAllocatedPct":40.27478236361195,"vramReservedGb":118.218555392,"vramReservedPct":41.12970616850285,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":71824.68422651701,"meanTps":71813.91680970397,"stepMs":912.4439697265625,"jitter":0.00021109186927011916,"achievedTflops":450.9938529922168,"nominalPeakTflops":2250.0,"mfuNominalPct":20.044171244098525,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.044171244098525,"vramAllocatedGb":217.571795456,"vramAllocatedPct":75.69593443251,"vramReservedGb":222.918868992,"vramReservedPct":77.55624783820012,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.98907904,"vramAllocatedPct":99.49915856535607,"vramReservedGb":286.273830912,"vramReservedPct":99.59822728420022,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 767.62 MiB is free. Including non-PyTorch memory, this process has 266.92 GiB memory in use. Of the allocated memory 265.91 GiB is allocated by PyTorch, and 199.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":66051.39763668606,"meanTps":66017.35796620033,"stepMs":496.0985107421875,"jitter":0.0002833657723264959,"achievedTflops":414.742851103163,"nominalPeakTflops":2250.0,"mfuNominalPct":18.43301560458502,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.43301560458502,"vramAllocatedGb":115.76125952,"vramAllocatedPct":40.27478236361195,"vramReservedGb":118.218555392,"vramReservedPct":41.12970616850285,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":128,"tokensPerStep":65536,"status":"complete","stable":true,"tps":71842.12396575826,"meanTps":71839.36807744013,"stepMs":912.2224731445312,"jitter":0.00023276830703473396,"achievedTflops":451.10335873220424,"nominalPeakTflops":2250.0,"mfuNominalPct":20.049038165875743,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.049038165875743,"vramAllocatedGb":217.571795456,"vramAllocatedPct":75.69593443251,"vramReservedGb":222.918868992,"vramReservedPct":77.55624783820012,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":256,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.98907904,"vramAllocatedPct":99.49915856535607,"vramReservedGb":286.273830912,"vramReservedPct":99.59822728420022,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 767.62 MiB is free. Including non-PyTorch memory, this process has 266.92 GiB memory in use. Of the allocated memory 265.91 GiB is allocated by PyTorch, and 199.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":2,"tokensPerStep":1024,"status":"complete","stable":true,"tps":8488.687235058049,"meanTps":8487.65659179678,"stepMs":120.63113784790039,"jitter":0.0007437208787388337,"achievedTflops":53.301254355828924,"nominalPeakTflops":165.2,"mfuNominalPct":32.264681813455766,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":25.935246625170187,"vramAllocatedGb":17.132302848,"vramAllocatedPct":67.82103014180717,"vramReservedGb":17.265852416,"vramReservedPct":68.34970800590474,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11505.654470826517,"meanTps":11504.288990162782,"stepMs":177.9994354248047,"jitter":0.0008557246743374041,"achievedTflops":72.24507141069272,"nominalPeakTflops":165.2,"mfuNominalPct":43.731883420516176,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":35.15290150548669,"vramAllocatedGb":20.313882112,"vramAllocatedPct":80.41583336684369,"vramReservedGb":20.57306112,"vramReservedPct":81.44183596962534,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.598305792,"vramAllocatedPct":97.37642705466035,"vramReservedGb":24.744296448,"vramReservedPct":97.95435500566866,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 14.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 15.69 MiB is free. Including non-PyTorch memory, this process has 23.50 GiB memory in use. Of the allocated memory 22.91 GiB is allocated by PyTorch, and 137.23 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":2,"tokensPerStep":1024,"status":"complete","stable":true,"tps":8460.063547359903,"meanTps":8457.20651635579,"stepMs":121.03927993774414,"jitter":0.0022042012219395182,"achievedTflops":53.121523566325955,"nominalPeakTflops":165.2,"mfuNominalPct":32.15588593603267,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":25.84779347968176,"vramAllocatedGb":17.132302848,"vramAllocatedPct":67.82103014180717,"vramReservedGb":17.265852416,"vramReservedPct":68.34970800590474,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11481.950140699471,"meanTps":11480.541179735474,"stepMs":178.36691284179688,"jitter":0.001590349848675567,"achievedTflops":72.09622972358024,"nominalPeakTflops":165.2,"mfuNominalPct":43.64178554696141,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":35.0804783343996,"vramAllocatedGb":20.313882112,"vramAllocatedPct":80.41583336684369,"vramReservedGb":20.57306112,"vramReservedPct":81.44183596962534,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.598305792,"vramAllocatedPct":97.37642705466035,"vramReservedGb":24.744296448,"vramReservedPct":97.95435500566866,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 14.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 15.69 MiB is free. Including non-PyTorch memory, this process has 23.50 GiB memory in use. Of the allocated memory 22.91 GiB is allocated by PyTorch, and 137.23 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15262.130042440062,"meanTps":15260.221856933591,"stepMs":134.18834686279297,"jitter":0.00029258622831552844,"achievedTflops":95.83232988537277,"nominalPeakTflops":209.5,"mfuNominalPct":45.743355553877215,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":35.63244557222733,"vramAllocatedGb":20.313882112,"vramAllocatedPct":60.33433684885138,"vramReservedGb":20.591935488,"vramReservedPct":61.1601842155462,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":18112.28375581875,"meanTps":18112.242693370357,"stepMs":226.14486694335938,"jitter":0.0002186291878280478,"achievedTflops":113.7287093635323,"nominalPeakTflops":209.5,"mfuNominalPct":54.28578012579108,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":42.28669021449162,"vramAllocatedGb":26.67704064,"vramAllocatedPct":79.23357767019111,"vramReservedGb":27.13714688,"vramReservedPct":80.60014092567144,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.843534336,"vramAllocatedPct":97.54870354221735,"vramReservedGb":33.013366784,"vramReservedPct":98.05312352796908,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 15.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.59 GiB is allocated by PyTorch, and 161.96 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15257.559281202419,"meanTps":15245.75798089951,"stepMs":134.22854614257812,"jitter":0.0004606157783248958,"achievedTflops":95.8036296516875,"nominalPeakTflops":209.5,"mfuNominalPct":45.72965615832339,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":35.62177423077165,"vramAllocatedGb":20.313882112,"vramAllocatedPct":60.33433684885138,"vramReservedGb":20.591935488,"vramReservedPct":61.1601842155462,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":18108.55711941974,"meanTps":18105.677235944735,"stepMs":226.19140625,"jitter":0.0003323520070589759,"achievedTflops":113.70530946798957,"nominalPeakTflops":209.5,"mfuNominalPct":54.274610724577364,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":42.27798964856236,"vramAllocatedGb":26.67704064,"vramAllocatedPct":79.23357767019111,"vramReservedGb":27.13714688,"vramReservedPct":80.60014092567144,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.843534336,"vramAllocatedPct":97.54870354221735,"vramReservedGb":33.013366784,"vramReservedPct":98.05312352796908,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 15.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.59 GiB is allocated by PyTorch, and 161.96 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":37549.42701498579,"meanTps":37545.3974604397,"stepMs":218.16577911376953,"jitter":0.0010949145466451365,"achievedTflops":235.77633441075943,"nominalPeakTflops":989.5,"mfuNominalPct":23.827825609980742,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.827825609980742,"vramAllocatedGb":39.4534272,"vramAllocatedPct":46.40624602528861,"vramReservedGb":40.093351936,"vramReservedPct":47.15894374622282,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":42074.168419035654,"meanTps":41999.4426264822,"stepMs":389.40757751464844,"jitter":0.0007247947401697047,"achievedTflops":264.18760529320735,"nominalPeakTflops":989.5,"mfuNominalPct":26.69910109077386,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.69910109077386,"vramAllocatedGb":64.906061312,"vramAllocatedPct":76.34436000979754,"vramReservedGb":66.192408576,"vramReservedPct":77.85739833989072,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.093110272,"vramAllocatedPct":98.91271408518236,"vramReservedGb":84.183875584,"vramReservedPct":99.01947483318739,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 44.19 MiB is free. Including non-PyTorch memory, this process has 79.13 GiB memory in use. Of the allocated memory 78.32 GiB is allocated by PyTorch, and 86.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":37861.61419103951,"meanTps":37852.13382026078,"stepMs":216.3668975830078,"jitter":0.0003705501218950507,"achievedTflops":237.7365866401907,"nominalPeakTflops":989.5,"mfuNominalPct":24.02593093887728,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.02593093887728,"vramAllocatedGb":39.4534272,"vramAllocatedPct":46.40624602528861,"vramReservedGb":40.093351936,"vramReservedPct":47.15894374622282,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":42069.267522536764,"meanTps":42036.556230725975,"stepMs":389.45294189453125,"jitter":0.0006379263123399788,"achievedTflops":264.15683210960134,"nominalPeakTflops":989.5,"mfuNominalPct":26.695991117695943,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.695991117695943,"vramAllocatedGb":64.906061312,"vramAllocatedPct":76.34436000979754,"vramReservedGb":66.192408576,"vramReservedPct":77.85739833989072,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.093110272,"vramAllocatedPct":98.91271408518236,"vramReservedGb":84.183875584,"vramReservedPct":99.01947483318739,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 44.19 MiB is free. Including non-PyTorch memory, this process has 79.13 GiB memory in use. Of the allocated memory 78.32 GiB is allocated by PyTorch, and 86.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":45831.31310496994,"meanTps":45808.433844201834,"stepMs":357.48484802246094,"jitter":0.005070859399046283,"achievedTflops":287.7790651987583,"nominalPeakTflops":989.5,"mfuNominalPct":29.083280970061473,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.083280970061473,"vramAllocatedGb":64.906061312,"vramAllocatedPct":43.23589996239077,"vramReservedGb":66.17563136,"vramReservedPct":44.08159915413062,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":47736.82911405461,"meanTps":47727.95444666214,"stepMs":686.43017578125,"jitter":0.001859565879859924,"achievedTflops":299.7439769297336,"nominalPeakTflops":989.5,"mfuNominalPct":30.29246861341421,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.29246861341421,"vramAllocatedGb":115.811329024,"vramAllocatedPct":77.14544581782289,"vramReservedGb":118.294052864,"vramReservedPct":78.79926361866728,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.075858432,"vramAllocatedPct":99.30395977951342,"vramReservedGb":149.258502144,"vramReservedPct":99.42562430669575,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 73.44 MiB is free. Including non-PyTorch memory, this process has 139.73 GiB memory in use. Of the allocated memory 138.84 GiB is allocated by PyTorch, and 174.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":45697.788044235545,"meanTps":45688.5591683282,"stepMs":358.5293884277344,"jitter":0.0031219895217100085,"achievedTflops":286.94064895982706,"nominalPeakTflops":989.5,"mfuNominalPct":28.998549667491364,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.998549667491364,"vramAllocatedGb":64.906061312,"vramAllocatedPct":43.23589996239077,"vramReservedGb":66.17563136,"vramReservedPct":44.08159915413062,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":47744.52582332644,"meanTps":47728.042065325164,"stepMs":686.3195190429688,"jitter":0.002161734588839021,"achievedTflops":299.79230528101345,"nominalPeakTflops":989.5,"mfuNominalPct":30.29735273178509,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.29735273178509,"vramAllocatedGb":115.811329024,"vramAllocatedPct":77.14544581782289,"vramReservedGb":118.294052864,"vramReservedPct":78.79926361866728,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.075858432,"vramAllocatedPct":99.30395977951342,"vramReservedGb":149.258502144,"vramReservedPct":99.42562430669575,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 73.44 MiB is free. Including non-PyTorch memory, this process has 139.73 GiB memory in use. Of the allocated memory 138.84 GiB is allocated by PyTorch, and 174.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":11370.24370745008,"meanTps":11351.837680452987,"stepMs":360.23854064941406,"jitter":0.0029008819712381846,"achievedTflops":71.3948146699995,"nominalPeakTflops":364.2,"mfuNominalPct":19.60318909115857,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.815133228132758,"vramAllocatedGb":26.67704064,"vramAllocatedPct":52.434545069423784,"vramReservedGb":27.072135168,"vramReservedPct":53.21111553369169,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":11452.92898342783,"meanTps":11459.452244193313,"stepMs":715.2755432128906,"jitter":0.004026044983330382,"achievedTflops":71.91400318576555,"nominalPeakTflops":364.2,"mfuNominalPct":19.74574497137989,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.93014207835962,"vramAllocatedGb":39.403357696,"vramAllocatedPct":77.44851323199613,"vramReservedGb":40.187723776,"vramReservedPct":78.99020892184785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":50.091461632,"vramAllocatedPct":98.45631072729122,"vramReservedGb":50.316967936,"vramReservedPct":98.89955031372412,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 108.50 MiB is free. Process 2319972 has 47.27 GiB memory in use. Of the allocated memory 46.60 GiB is allocated by PyTorch, and 171.06 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7787.911763357667,"meanTps":7785.667785048263,"stepMs":525.9432983398438,"jitter":0.0007418974580062446,"achievedTflops":48.90102018190808,"nominalPeakTflops":154.8,"mfuNominalPct":31.589806319062063,"configuredPeakTflops":180.6,"mfuConfiguredPct":27.076976844910345,"vramAllocatedGb":26.67704064,"vramAllocatedPct":52.27314398798792,"vramReservedGb":27.051163648,"vramReservedPct":53.006230769625894,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":8861.635454325293,"meanTps":8860.464476618565,"stepMs":924.434326171875,"jitter":0.0004931580836359228,"achievedTflops":55.64303081033401,"nominalPeakTflops":154.8,"mfuNominalPct":35.945110342593026,"configuredPeakTflops":180.6,"mfuConfiguredPct":30.810094579365455,"vramAllocatedGb":39.403357696,"vramAllocatedPct":77.21011555400172,"vramReservedGb":40.166752256,"vramReservedPct":78.70597239558452,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.559065088,"vramAllocatedPct":99.06950793036229,"vramReservedGb":50.67767808,"vramReservedPct":99.3019277888216,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 2.25 MiB is free. Process 1415479 has 47.52 GiB memory in use. Of the allocated memory 47.09 GiB is allocated by PyTorch, and 113.12 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7764.621651385148,"meanTps":7772.3928283863615,"stepMs":527.5208740234375,"jitter":0.0014929091872151275,"achievedTflops":48.754779409000804,"nominalPeakTflops":154.8,"mfuNominalPct":31.49533553553023,"configuredPeakTflops":180.6,"mfuConfiguredPct":26.996001887597345,"vramAllocatedGb":26.67704064,"vramAllocatedPct":52.27314398798792,"vramReservedGb":27.051163648,"vramReservedPct":53.006230769625894,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":8862.03215872114,"meanTps":8861.935436808935,"stepMs":924.3929443359375,"jitter":0.0009002153149758993,"achievedTflops":55.645521754024315,"nominalPeakTflops":154.8,"mfuNominalPct":35.94671947934387,"configuredPeakTflops":180.6,"mfuConfiguredPct":30.811473839437603,"vramAllocatedGb":39.403357696,"vramAllocatedPct":77.21011555400172,"vramReservedGb":40.166752256,"vramReservedPct":78.70597239558452,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.559065088,"vramAllocatedPct":99.06950793036229,"vramReservedGb":50.67767808,"vramReservedPct":99.3019277888216,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 2.25 MiB is free. Process 1449495 has 47.52 GiB memory in use. Of the allocated memory 47.09 GiB is allocated by PyTorch, and 113.12 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22552.328079589795,"meanTps":22568.17339588269,"stepMs":181.6220474243164,"jitter":0.0007540185750684765,"achievedTflops":141.60815942444108,"nominalPeakTflops":468.0,"mfuNominalPct":30.258153723171173,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.258153723171173,"vramAllocatedGb":26.67704064,"vramAllocatedPct":26.160642515676425,"vramReservedGb":27.051163648,"vramReservedPct":26.527523475272158,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25559.920026299245,"meanTps":25555.699231841834,"stepMs":320.5017852783203,"jitter":0.0006326550930626218,"achievedTflops":160.49310816987654,"nominalPeakTflops":468.0,"mfuNominalPct":34.29339918159755,"configuredPeakTflops":468.0,"mfuConfiguredPct":34.29339918159755,"vramAllocatedGb":39.403357696,"vramAllocatedPct":38.64061117246862,"vramReservedGb":40.166752256,"vramReservedPct":39.389228399247045,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25199.200735111164,"meanTps":25197.410607776917,"stepMs":650.1793518066406,"jitter":0.0011589359850706007,"achievedTflops":158.22811828884238,"nominalPeakTflops":468.0,"mfuNominalPct":33.80942698479538,"configuredPeakTflops":468.0,"mfuConfiguredPct":33.80942698479538,"vramAllocatedGb":64.855991808,"vramAllocatedPct":63.600548486053015,"vramReservedGb":66.181922816,"vramReservedPct":64.90081291826023,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22584.777094387777,"meanTps":22579.68617506965,"stepMs":181.36109924316406,"jitter":0.0007426296482296638,"achievedTflops":141.81190979754948,"nominalPeakTflops":468.0,"mfuNominalPct":30.30169012768151,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.30169012768151,"vramAllocatedGb":26.727179776,"vramAllocatedPct":26.20981108840687,"vramReservedGb":28.533850112,"vramReservedPct":27.98150898554562,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25548.6906907167,"meanTps":25545.683119012836,"stepMs":320.6426544189453,"jitter":0.00039489176480968046,"achievedTflops":160.42259813039007,"nominalPeakTflops":468.0,"mfuNominalPct":34.278332933844034,"configuredPeakTflops":468.0,"mfuConfiguredPct":34.278332933844034,"vramAllocatedGb":39.450125824,"vramAllocatedPct":38.68647399114652,"vramReservedGb":40.596668416,"vramReservedPct":39.8108225005286,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25197.27295259657,"meanTps":25195.932785360703,"stepMs":650.2290954589844,"jitter":0.0003200155425237696,"achievedTflops":158.2160135636584,"nominalPeakTflops":468.0,"mfuNominalPct":33.80684050505522,"configuredPeakTflops":468.0,"mfuConfiguredPct":33.80684050505522,"vramAllocatedGb":64.886793728,"vramAllocatedPct":63.63075416099239,"vramReservedGb":66.651684864,"vramReservedPct":65.36148159478252,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":16588.336022196265,"meanTps":16584.65990977172,"stepMs":493.8409729003906,"jitter":0.0023957529210679326,"achievedTflops":109.27357474294598,"nominalPeakTflops":312.0,"mfuNominalPct":35.02358164838012,"configuredPeakTflops":312.0,"mfuConfiguredPct":35.02358164838012,"vramAllocatedGb":39.407752704,"vramAllocatedPct":46.376116835787954,"vramReservedGb":40.074477568,"vramReservedPct":47.16073681710067,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":17783.633500165375,"meanTps":17810.083402083405,"stepMs":921.2965393066406,"jitter":0.0025274243156747753,"achievedTflops":117.14744636720908,"nominalPeakTflops":312.0,"mfuNominalPct":37.547258451028554,"configuredPeakTflops":312.0,"mfuConfiguredPct":37.547258451028554,"vramAllocatedGb":64.862946816,"vramAllocatedPct":76.33248265758088,"vramReservedGb":66.194505728,"vramReservedPct":77.89949745643234,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.318431232,"vramAllocatedPct":99.22822667907857,"vramReservedGb":84.395687936,"vramReservedPct":99.319144472792,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 26.75 MiB is free. Process 764857 has 79.10 GiB memory in use. Of the allocated memory 78.53 GiB is allocated by PyTorch, and 73.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":16578.878729044754,"meanTps":16579.70738612557,"stepMs":494.1226806640625,"jitter":0.002016993422191962,"achievedTflops":109.21127601517277,"nominalPeakTflops":312.0,"mfuNominalPct":35.00361410742717,"configuredPeakTflops":312.0,"mfuConfiguredPct":35.00361410742717,"vramAllocatedGb":39.407752704,"vramAllocatedPct":46.376116835787954,"vramReservedGb":40.074477568,"vramReservedPct":47.16073681710067,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":17765.490235631674,"meanTps":17778.67333484613,"stepMs":922.2374267578125,"jitter":0.0034000347254899387,"achievedTflops":117.02792989669318,"nominalPeakTflops":312.0,"mfuNominalPct":37.508951889965765,"configuredPeakTflops":312.0,"mfuConfiguredPct":37.508951889965765,"vramAllocatedGb":64.862946816,"vramAllocatedPct":76.33248265758088,"vramReservedGb":66.194505728,"vramReservedPct":77.89949745643234,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.318431232,"vramAllocatedPct":99.22822667907857,"vramReservedGb":84.395687936,"vramReservedPct":99.319144472792,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 26.75 MiB is free. Process 852916 has 79.10 GiB memory in use. Of the allocated memory 78.53 GiB is allocated by PyTorch, and 73.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":60352.920372884124,"meanTps":60216.88225156986,"stepMs":271.4698791503906,"jitter":0.0022361855202475403,"achievedTflops":397.56726331664083,"nominalPeakTflops":2250.0,"mfuNominalPct":17.669656147406258,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.669656147406258,"vramAllocatedGb":64.862946816,"vramAllocatedPct":33.8704585270579,"vramReservedGb":66.135785472,"vramReservedPct":34.53511579327772,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":65827.5190000051,"meanTps":65778.66683236585,"stepMs":497.7857360839844,"jitter":0.0007248736362564612,"achievedTflops":433.63049241134104,"nominalPeakTflops":2250.0,"mfuNominalPct":19.272466329392934,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.272466329392934,"vramAllocatedGb":115.77333504,"vramAllocatedPct":60.45510010723438,"vramReservedGb":118.260498432,"vramReservedPct":61.7538595477779,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.336147456,"vramAllocatedPct":99.39068304892639,"vramReservedGb":190.595465216,"vramReservedPct":99.52609489600644,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 11.81 MiB is free. Including non-PyTorch memory, this process has 178.33 GiB memory in use. Of the allocated memory 177.26 GiB is allocated by PyTorch, and 247.30 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":60037.56795200134,"meanTps":59685.15470123575,"stepMs":272.8957977294922,"jitter":0.0032163000372557263,"achievedTflops":395.4899189532523,"nominalPeakTflops":2250.0,"mfuNominalPct":17.577329731255656,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.577329731255656,"vramAllocatedGb":64.862946816,"vramAllocatedPct":33.8704585270579,"vramReservedGb":66.135785472,"vramReservedPct":34.53511579327772,"warmupSteps":15,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":65833.8899136777,"meanTps":65561.26153356924,"stepMs":497.73756408691406,"jitter":0.001847174959669346,"achievedTflops":433.6724600029339,"nominalPeakTflops":2250.0,"mfuNominalPct":19.274331555685954,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.274331555685954,"vramAllocatedGb":115.77333504,"vramAllocatedPct":60.45510010723438,"vramReservedGb":118.260498432,"vramReservedPct":61.7538595477779,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.336147456,"vramAllocatedPct":99.39068304892639,"vramReservedGb":190.595465216,"vramReservedPct":99.52609489600644,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 11.81 MiB is free. Including non-PyTorch memory, this process has 178.33 GiB memory in use. Of the allocated memory 177.26 GiB is allocated by PyTorch, and 247.30 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":63745.216193064545,"meanTps":63690.76268351657,"stepMs":514.0464172363281,"jitter":0.00032073776054330154,"achievedTflops":419.91358487418967,"nominalPeakTflops":2250.0,"mfuNominalPct":18.66282599440843,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.66282599440843,"vramAllocatedGb":115.77333504,"vramAllocatedPct":40.27898358725053,"vramReservedGb":118.241624064,"vramReservedPct":41.13773204648716,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":69102.97502881028,"meanTps":69101.70046437533,"stepMs":948.3817443847656,"jitter":0.00024574371826648967,"achievedTflops":455.2071465556721,"nominalPeakTflops":2250.0,"mfuNominalPct":20.23142873580765,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.23142873580765,"vramAllocatedGb":217.594110976,"vramAllocatedPct":75.70369827954362,"vramReservedGb":222.960812032,"vramReservedPct":77.57084034362614,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.990914048,"vramAllocatedPct":99.49979698746846,"vramReservedGb":286.273830912,"vramReservedPct":99.59822728420022,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 767.62 MiB is free. Including non-PyTorch memory, this process has 266.92 GiB memory in use. Of the allocated memory 265.91 GiB is allocated by PyTorch, and 197.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":63772.05120689711,"meanTps":63768.930021542736,"stepMs":513.8301086425781,"jitter":0.00040522956776919657,"achievedTflops":420.09035714874676,"nominalPeakTflops":2250.0,"mfuNominalPct":18.670682539944302,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.670682539944302,"vramAllocatedGb":115.77333504,"vramAllocatedPct":40.27898358725053,"vramReservedGb":118.241624064,"vramReservedPct":41.13773204648716,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":64,"tokensPerStep":65536,"status":"complete","stable":true,"tps":69131.46042038429,"meanTps":69131.49669555206,"stepMs":947.990966796875,"jitter":0.00018451501440520674,"achievedTflops":455.3947904857854,"nominalPeakTflops":2250.0,"mfuNominalPct":20.239768466034906,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.239768466034906,"vramAllocatedGb":217.594110976,"vramAllocatedPct":75.70369827954362,"vramReservedGb":222.960812032,"vramReservedPct":77.57084034362614,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.990914048,"vramAllocatedPct":99.49979698746846,"vramReservedGb":286.273830912,"vramReservedPct":99.59822728420022,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 767.62 MiB is free. Including non-PyTorch memory, this process has 266.92 GiB memory in use. Of the allocated memory 265.91 GiB is allocated by PyTorch, and 197.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":1,"tokensPerStep":1024,"status":"complete","stable":true,"tps":8355.657724876255,"meanTps":8357.768122924688,"stepMs":122.55169296264648,"jitter":0.0010065154004718455,"achievedTflops":55.04184311820161,"nominalPeakTflops":165.2,"mfuNominalPct":33.31830697227701,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":26.782179767189174,"vramAllocatedGb":17.134449664,"vramAllocatedPct":67.82952866497342,"vramReservedGb":17.226006528,"vramReservedPct":68.19197152441413,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11291.546836020863,"meanTps":11267.729370667814,"stepMs":181.37461853027344,"jitter":0.0013415231075549398,"achievedTflops":74.38164295070985,"nominalPeakTflops":165.2,"mfuNominalPct":45.02520759728199,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":36.192511370064,"vramAllocatedGb":20.31635712,"vramAllocatedPct":80.42563109185815,"vramReservedGb":20.575158272,"vramReservedPct":81.4501378897038,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.6001408,"vramAllocatedPct":97.38369123472901,"vramReservedGb":24.7463936,"vramReservedPct":97.96265692574711,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 14.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 13.69 MiB is free. Including non-PyTorch memory, this process has 23.50 GiB memory in use. Of the allocated memory 22.91 GiB is allocated by PyTorch, and 137.48 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":1,"tokensPerStep":1024,"status":"complete","stable":true,"tps":8345.157632853667,"meanTps":8340.914520445303,"stepMs":122.70589065551758,"jitter":0.0014947398409649161,"achievedTflops":54.9726750841744,"nominalPeakTflops":165.2,"mfuNominalPct":33.276437702284746,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":26.74852408604707,"vramAllocatedGb":17.134449664,"vramAllocatedPct":67.82952866497342,"vramReservedGb":17.226006528,"vramReservedPct":68.19197152441413,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11286.108664732445,"meanTps":11284.591516872815,"stepMs":181.4620132446289,"jitter":0.0008793525168755281,"achievedTflops":74.34581968212193,"nominalPeakTflops":165.2,"mfuNominalPct":45.00352281000117,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":36.17508053626889,"vramAllocatedGb":20.31635712,"vramAllocatedPct":80.42563109185815,"vramReservedGb":20.575158272,"vramReservedPct":81.4501378897038,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.6001408,"vramAllocatedPct":97.38369123472901,"vramReservedGb":24.7463936,"vramReservedPct":97.96265692574711,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 14.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 13.69 MiB is free. Including non-PyTorch memory, this process has 23.50 GiB memory in use. Of the allocated memory 22.91 GiB is allocated by PyTorch, and 137.48 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":14983.423902046805,"meanTps":14980.923248228764,"stepMs":136.68437957763672,"jitter":0.0004232927418416293,"achievedTflops":98.70141824199558,"nominalPeakTflops":209.5,"mfuNominalPct":47.112848802861855,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":36.699232060999506,"vramAllocatedGb":20.31635712,"vramAllocatedPct":60.34168787932169,"vramReservedGb":20.59403264,"vramReservedPct":61.16641297450491,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":17782.305822614657,"meanTps":17778.64684126334,"stepMs":230.3413314819336,"jitter":0.0006814559453039765,"achievedTflops":117.13870045852487,"nominalPeakTflops":209.5,"mfuNominalPct":55.91346083939135,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":43.55459554705986,"vramAllocatedGb":26.680155648,"vramAllocatedPct":79.24282956704286,"vramReservedGb":27.139244032,"vramReservedPct":80.60636968463015,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.845369344,"vramAllocatedPct":97.55415370630622,"vramReservedGb":33.015463936,"vramReservedPct":98.05935228692778,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 13.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.59 GiB is allocated by PyTorch, and 162.21 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":14978.272154187529,"meanTps":14976.109760429683,"stepMs":136.73139190673828,"jitter":0.0003274323797180138,"achievedTflops":98.66748175168068,"nominalPeakTflops":209.5,"mfuNominalPct":47.096650000802235,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":36.686613770851444,"vramAllocatedGb":20.31635712,"vramAllocatedPct":60.34168787932169,"vramReservedGb":20.59403264,"vramReservedPct":61.16641297450491,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":17793.520149676504,"meanTps":17791.915695592947,"stepMs":230.19615936279297,"jitter":0.00039374145220954696,"achievedTflops":117.21257342593672,"nominalPeakTflops":209.5,"mfuNominalPct":55.94872239901514,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":43.58206304674092,"vramAllocatedGb":26.680155648,"vramAllocatedPct":79.24282956704286,"vramReservedGb":27.139244032,"vramReservedPct":80.60636968463015,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.845369344,"vramAllocatedPct":97.55415370630622,"vramReservedGb":33.015463936,"vramReservedPct":98.05935228692778,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 13.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.59 GiB is allocated by PyTorch, and 162.21 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":36993.8904060184,"meanTps":36991.98822365945,"stepMs":221.4419708251953,"jitter":0.0003345254179161423,"achievedTflops":243.6925947789663,"nominalPeakTflops":989.5,"mfuNominalPct":24.627851923089064,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.627851923089064,"vramAllocatedGb":39.457822208,"vramAllocatedPct":46.411415559014976,"vramReservedGb":40.116420608,"vramReservedPct":47.18607777599625,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":41039.08521529468,"meanTps":41026.413290652396,"stepMs":399.2291717529297,"jitter":0.0004935033305262831,"achievedTflops":270.33980621414344,"nominalPeakTflops":989.5,"mfuNominalPct":27.320849541601156,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.320849541601156,"vramAllocatedGb":64.91301632,"vramAllocatedPct":76.3525406885183,"vramReservedGb":66.215477248,"vramReservedPct":77.88453236966416,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.09494528,"vramAllocatedPct":98.91487247391433,"vramReservedGb":84.185972736,"vramReservedPct":99.02194156316679,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 42.19 MiB is free. Including non-PyTorch memory, this process has 79.13 GiB memory in use. Of the allocated memory 78.32 GiB is allocated by PyTorch, and 86.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":36981.685427789314,"meanTps":36970.05213948104,"stepMs":221.51505279541016,"jitter":0.00043087405807059145,"achievedTflops":243.61219602173304,"nominalPeakTflops":989.5,"mfuNominalPct":24.619726732868422,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.619726732868422,"vramAllocatedGb":39.457822208,"vramAllocatedPct":46.411415559014976,"vramReservedGb":40.116420608,"vramReservedPct":47.18607777599625,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":41047.899163037,"meanTps":41029.12521104512,"stepMs":399.14344787597656,"jitter":0.0002038143431913393,"achievedTflops":270.3978669850436,"nominalPeakTflops":989.5,"mfuNominalPct":27.3267172294132,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.3267172294132,"vramAllocatedGb":64.91301632,"vramAllocatedPct":76.3525406885183,"vramReservedGb":66.215477248,"vramReservedPct":77.88453236966416,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.09494528,"vramAllocatedPct":98.91487247391433,"vramReservedGb":84.185972736,"vramReservedPct":99.02194156316679,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 42.19 MiB is free. Including non-PyTorch memory, this process has 79.13 GiB memory in use. Of the allocated memory 78.32 GiB is allocated by PyTorch, and 86.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":44493.11970561512,"meanTps":44501.99179376608,"stepMs":368.23670959472656,"jitter":0.0037673882321214718,"achievedTflops":293.0928234871063,"nominalPeakTflops":989.5,"mfuNominalPct":29.62029545094556,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.62029545094556,"vramAllocatedGb":64.91301632,"vramAllocatedPct":43.24053290458518,"vramReservedGb":66.198700032,"vramReservedPct":44.09696589761645,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":46371.90468688242,"meanTps":46427.73544809218,"stepMs":706.634765625,"jitter":0.0023576885484125074,"achievedTflops":305.46908297459976,"nominalPeakTflops":989.5,"mfuNominalPct":30.871054368327414,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.871054368327414,"vramAllocatedGb":115.823404544,"vramAllocatedPct":77.15348968867501,"vramReservedGb":118.323412992,"vramReservedPct":78.8188212921947,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.077697536,"vramAllocatedPct":99.30518486257897,"vramReservedGb":149.268987904,"vramReservedPct":99.4326091900984,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 63.44 MiB is free. Including non-PyTorch memory, this process has 139.74 GiB memory in use. Of the allocated memory 138.84 GiB is allocated by PyTorch, and 182.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":44456.420183215254,"meanTps":44509.46311967961,"stepMs":368.5406951904297,"jitter":0.002457896151744301,"achievedTflops":292.851070004501,"nominalPeakTflops":989.5,"mfuNominalPct":29.59586356791319,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.59586356791319,"vramAllocatedGb":64.91301632,"vramAllocatedPct":43.24053290458518,"vramReservedGb":66.198700032,"vramReservedPct":44.09696589761645,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":46427.55959192902,"meanTps":46436.83119587928,"stepMs":705.7876892089844,"jitter":0.0030282483119156703,"achievedTflops":305.83570265352864,"nominalPeakTflops":989.5,"mfuNominalPct":30.908105371756307,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.908105371756307,"vramAllocatedGb":115.823404544,"vramAllocatedPct":77.15348968867501,"vramReservedGb":118.323412992,"vramReservedPct":78.8188212921947,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.077697536,"vramAllocatedPct":99.30518486257897,"vramReservedGb":149.268987904,"vramReservedPct":99.4326091900984,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 63.44 MiB is free. Including non-PyTorch memory, this process has 139.74 GiB memory in use. Of the allocated memory 138.84 GiB is allocated by PyTorch, and 182.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":11192.486847229391,"meanTps":11188.749805895204,"stepMs":365.95977783203125,"jitter":0.002424169201043716,"achievedTflops":73.72909774817981,"nominalPeakTflops":364.2,"mfuNominalPct":20.24412348934097,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.33221556869537,"vramAllocatedGb":26.680155648,"vramAllocatedPct":52.44066771359454,"vramReservedGb":27.07423232,"vramReservedPct":53.21523755054301,"warmupSteps":13,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":11293.910310849065,"meanTps":11287.792252496594,"stepMs":725.3466491699219,"jitter":0.0023863163150330655,"achievedTflops":74.39721204353188,"nominalPeakTflops":364.2,"mfuNominalPct":20.427570577576024,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.480213944228,"vramAllocatedGb":39.407752704,"vramAllocatedPct":77.45715175559273,"vramReservedGb":40.187723776,"vramReservedPct":78.99020892184785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":50.09329664,"vramAllocatedPct":98.45991749203614,"vramReservedGb":50.319065088,"vramReservedPct":98.90367233057545,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 106.50 MiB is free. Process 2319972 has 47.27 GiB memory in use. Of the allocated memory 46.60 GiB is allocated by PyTorch, and 171.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7682.265023283575,"meanTps":7682.053659360586,"stepMs":533.1760864257812,"jitter":0.0011558790469800607,"achievedTflops":50.60595349006882,"nominalPeakTflops":154.8,"mfuNominalPct":32.69118442510906,"configuredPeakTflops":180.6,"mfuConfiguredPct":28.021015221522052,"vramAllocatedGb":26.680155648,"vramAllocatedPct":52.279247785778125,"vramReservedGb":27.0532608,"vramReservedPct":53.01034009831569,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":8724.85916446521,"meanTps":8723.989820422612,"stepMs":938.9263305664062,"jitter":0.00033876448000434755,"achievedTflops":57.47391111164857,"nominalPeakTflops":154.8,"mfuNominalPct":37.12784955532853,"configuredPeakTflops":180.6,"mfuConfiguredPct":31.823871047424458,"vramAllocatedGb":39.407752704,"vramAllocatedPct":77.21872748697857,"vramReservedGb":40.166752256,"vramReservedPct":78.70597239558452,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.560904192,"vramAllocatedPct":99.07311161899845,"vramReservedGb":50.679775232,"vramReservedPct":99.3060371175114,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 256.00 KiB is free. Process 1415479 has 47.52 GiB memory in use. Of the allocated memory 47.09 GiB is allocated by PyTorch, and 113.36 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7666.7617745078205,"meanTps":7668.6703516369425,"stepMs":534.2542419433594,"jitter":0.002544306219144574,"achievedTflops":50.503827790927616,"nominalPeakTflops":154.8,"mfuNominalPct":32.62521175124523,"configuredPeakTflops":180.6,"mfuConfiguredPct":27.964467215353057,"vramAllocatedGb":26.680155648,"vramAllocatedPct":52.279247785778125,"vramReservedGb":27.0532608,"vramReservedPct":53.01034009831569,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":8734.569276501272,"meanTps":8735.502729955897,"stepMs":937.8825378417969,"jitter":0.0005021513324559786,"achievedTflops":57.53787525198883,"nominalPeakTflops":154.8,"mfuNominalPct":37.1691700594243,"configuredPeakTflops":180.6,"mfuConfiguredPct":31.859288622363692,"vramAllocatedGb":39.407752704,"vramAllocatedPct":77.21872748697857,"vramReservedGb":40.166752256,"vramReservedPct":78.70597239558452,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.560904192,"vramAllocatedPct":99.07311161899845,"vramReservedGb":50.679775232,"vramReservedPct":99.3060371175114,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 256.00 KiB is free. Process 1449495 has 47.52 GiB memory in use. Of the allocated memory 47.09 GiB is allocated by PyTorch, and 113.36 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22391.97225363784,"meanTps":22383.869598994326,"stepMs":182.92269897460938,"jitter":0.0004255603301361976,"achievedTflops":147.5042976236931,"nominalPeakTflops":468.0,"mfuNominalPct":31.518012312754937,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.518012312754937,"vramAllocatedGb":26.680155648,"vramAllocatedPct":26.163697225223153,"vramReservedGb":27.0532608,"vramReservedPct":26.529580031863773,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25269.31883846712,"meanTps":25269.082512740646,"stepMs":324.18760681152344,"jitter":0.00017822496974601146,"achievedTflops":166.458455935774,"nominalPeakTflops":468.0,"mfuNominalPct":35.56804614012265,"configuredPeakTflops":468.0,"mfuConfiguredPct":35.56804614012265,"vramAllocatedGb":39.407752704,"vramAllocatedPct":38.64492110454441,"vramReservedGb":40.166752256,"vramReservedPct":39.389228399247045,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":24979.28765671118,"meanTps":24973.945934576037,"stepMs":655.9034118652344,"jitter":0.00023046800712225106,"achievedTflops":164.5479120466832,"nominalPeakTflops":468.0,"mfuNominalPct":35.1598102663853,"configuredPeakTflops":468.0,"mfuConfiguredPct":35.1598102663853,"vramAllocatedGb":64.862946816,"vramAllocatedPct":63.60736886318693,"vramReservedGb":66.184019968,"vramReservedPct":64.90286947485185,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22244.802963305854,"meanTps":22196.8670019092,"stepMs":184.13289642333984,"jitter":0.00048456226058557174,"achievedTflops":146.53483845518815,"nominalPeakTflops":468.0,"mfuNominalPct":31.310862917775246,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.310862917775246,"vramAllocatedGb":26.729014784,"vramAllocatedPct":26.211610575424533,"vramReservedGb":28.535947264,"vramReservedPct":27.98356554213724,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25235.722218848296,"meanTps":25226.281621863476,"stepMs":324.61920166015625,"jitter":0.0003259396761642858,"achievedTflops":166.23714243452133,"nominalPeakTflops":468.0,"mfuNominalPct":35.520756930453274,"configuredPeakTflops":468.0,"mfuConfiguredPct":35.520756930453274,"vramAllocatedGb":39.456056832,"vramAllocatedPct":38.69229019025719,"vramReservedGb":40.598765568,"vramReservedPct":39.81287905712021,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":24952.672037289463,"meanTps":24948.77449571236,"stepMs":656.60302734375,"jitter":0.0003773415417152826,"achievedTflops":164.3725850051814,"nominalPeakTflops":468.0,"mfuNominalPct":35.122347223329356,"configuredPeakTflops":468.0,"mfuConfiguredPct":35.122347223329356,"vramAllocatedGb":64.894772736,"vramAllocatedPct":63.63857871614956,"vramReservedGb":66.660073472,"vramReservedPct":65.36970782114899,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11761.1212914424,"meanTps":11738.743945117969,"stepMs":174.133056640625,"jitter":0.0026860795020392596,"achievedTflops":84.72637410432955,"nominalPeakTflops":312.0,"mfuNominalPct":27.15588913600306,"configuredPeakTflops":312.0,"mfuConfiguredPct":27.15588913600306,"vramAllocatedGb":20.320330752,"vramAllocatedPct":23.913518747821232,"vramReservedGb":20.451426304,"vramReservedPct":24.067795564412883,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":14194.49709098835,"meanTps":14201.09855641411,"stepMs":288.5625305175781,"jitter":0.003223876199723402,"achievedTflops":102.25625949704002,"nominalPeakTflops":312.0,"mfuNominalPct":32.77444214648719,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.77444214648719,"vramAllocatedGb":26.684465664,"vramAllocatedPct":31.40300606912365,"vramReservedGb":27.145535488,"vramReservedPct":31.94560559739134,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":15916.067234723278,"meanTps":15920.85688968388,"stepMs":514.7000122070312,"jitter":0.0006650109779084045,"achievedTflops":114.6583419541831,"nominalPeakTflops":312.0,"mfuNominalPct":36.749468575058685,"configuredPeakTflops":312.0,"mfuConfiguredPct":36.749468575058685,"vramAllocatedGb":39.41270272,"vramAllocatedPct":46.381942149453494,"vramReservedGb":40.187723776,"vramReservedPct":47.29400803946309,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":17064.778312260154,"meanTps":17066.85264959616,"stepMs":960.1062316894531,"jitter":0.0008404338471024271,"achievedTflops":122.93358392145988,"nominalPeakTflops":312.0,"mfuNominalPct":39.40178971841663,"configuredPeakTflops":312.0,"mfuConfiguredPct":39.40178971841663,"vramAllocatedGb":64.869176832,"vramAllocatedPct":76.33981431011318,"vramReservedGb":66.339209216,"vramReservedPct":78.06978846278432,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.183681024,"vramAllocatedPct":99.06964896375455,"vramReservedGb":84.37891072,"vramReservedPct":99.29940058799757,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 102.75 MiB is free. Process 703773 has 79.03 GiB memory in use. Of the allocated memory 78.40 GiB is allocated by PyTorch, and 126.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11758.431026243157,"meanTps":11748.28571538986,"stepMs":174.1728973388672,"jitter":0.003936335507304464,"achievedTflops":84.70699360394505,"nominalPeakTflops":312.0,"mfuNominalPct":27.149677437161873,"configuredPeakTflops":312.0,"mfuConfiguredPct":27.149677437161873,"vramAllocatedGb":20.320330752,"vramAllocatedPct":23.913518747821232,"vramReservedGb":20.451426304,"vramReservedPct":24.067795564412883,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":14169.010270700557,"meanTps":14170.280530325233,"stepMs":289.0815887451172,"jitter":0.002673418325257343,"achievedTflops":102.07265405526938,"nominalPeakTflops":312.0,"mfuNominalPct":32.715594248483775,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.715594248483775,"vramAllocatedGb":26.684465664,"vramAllocatedPct":31.40300606912365,"vramReservedGb":27.145535488,"vramReservedPct":31.94560559739134,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":15891.78625718477,"meanTps":15895.947078367371,"stepMs":515.4864196777344,"jitter":0.0020334178474458596,"achievedTflops":114.48342332733048,"nominalPeakTflops":312.0,"mfuNominalPct":36.69340491260593,"configuredPeakTflops":312.0,"mfuConfiguredPct":36.69340491260593,"vramAllocatedGb":39.41270272,"vramAllocatedPct":46.381942149453494,"vramReservedGb":40.187723776,"vramReservedPct":47.29400803946309,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":17023.259014278658,"meanTps":17027.166179512362,"stepMs":962.4479064941406,"jitter":0.001484504262910181,"achievedTflops":122.63448152414945,"nominalPeakTflops":312.0,"mfuNominalPct":39.305923565432515,"configuredPeakTflops":312.0,"mfuConfiguredPct":39.305923565432515,"vramAllocatedGb":64.869176832,"vramAllocatedPct":76.33981431011318,"vramReservedGb":66.339209216,"vramReservedPct":78.06978846278432,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.183681024,"vramAllocatedPct":99.06964896375455,"vramReservedGb":84.37891072,"vramReservedPct":99.29940058799757,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 102.75 MiB is free. Process 736416 has 79.03 GiB memory in use. Of the allocated memory 78.40 GiB is allocated by PyTorch, and 126.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11264.633738778608,"meanTps":11255.193896177021,"stepMs":181.80795288085938,"jitter":0.02800734582037414,"achievedTflops":81.14970916884052,"nominalPeakTflops":2250.0,"mfuNominalPct":3.6066537408373565,"configuredPeakTflops":2250.0,"mfuConfiguredPct":3.6066537408373565,"vramAllocatedGb":20.320330752,"vramAllocatedPct":10.610972115468854,"vramReservedGb":20.451426304,"vramReservedPct":10.679428247591462,"warmupSteps":21,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":21550.588522704387,"meanTps":21314.68188223833,"stepMs":190.06441497802734,"jitter":0.05312986800041088,"achievedTflops":155.24907703074894,"nominalPeakTflops":2250.0,"mfuNominalPct":6.899958979144397,"configuredPeakTflops":2250.0,"mfuConfiguredPct":6.899958979144397,"vramAllocatedGb":26.684465664,"vramAllocatedPct":13.934227967673293,"vramReservedGb":27.145535488,"vramReservedPct":14.174991718296132,"warmupSteps":24,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":42914.38017374511,"meanTps":42908.28071465955,"stepMs":190.8917236328125,"jitter":0.057389260807911575,"achievedTflops":309.1524812095726,"nominalPeakTflops":2250.0,"mfuNominalPct":13.740110275981005,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.740110275981005,"vramAllocatedGb":39.41270272,"vramAllocatedPct":20.580722561123764,"vramReservedGb":40.187723776,"vramReservedPct":20.985426938945363,"warmupSteps":13,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":55950.72900902857,"meanTps":55914.839616167825,"stepMs":292.8290710449219,"jitter":0.0014777447956176605,"achievedTflops":403.0655139045451,"nominalPeakTflops":2250.0,"mfuNominalPct":17.914022840202005,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.914022840202005,"vramAllocatedGb":64.869176832,"vramAllocatedPct":33.87371174802471,"vramReservedGb":66.339209216,"vramReservedPct":34.64134062305791,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":61181.148689585265,"meanTps":61160.05725661616,"stepMs":535.5898132324219,"jitter":0.0007936673909316283,"achievedTflops":440.7451265533783,"nominalPeakTflops":2250.0,"mfuNominalPct":19.58867229126126,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.58867229126126,"vramAllocatedGb":115.782125056,"vramAllocatedPct":60.459690121826604,"vramReservedGb":118.277275648,"vramReservedPct":61.76262035848142,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.339817472,"vramAllocatedPct":99.39259947626779,"vramReservedGb":190.587076608,"vramReservedPct":99.52171449065467,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 19.81 MiB is free. Including non-PyTorch memory, this process has 178.32 GiB memory in use. Of the allocated memory 177.27 GiB is allocated by PyTorch, and 235.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11551.341469785484,"meanTps":11500.303130484428,"stepMs":177.29542541503906,"jitter":0.02583084260381042,"achievedTflops":83.21513353391082,"nominalPeakTflops":2250.0,"mfuNominalPct":3.698450379284925,"configuredPeakTflops":2250.0,"mfuConfiguredPct":3.698450379284925,"vramAllocatedGb":20.320330752,"vramAllocatedPct":10.610972115468854,"vramReservedGb":20.451426304,"vramReservedPct":10.679428247591462,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":23341.846184855698,"meanTps":23272.812722259765,"stepMs":175.4788360595703,"jitter":0.011024375625044509,"achievedTflops":168.1531839640825,"nominalPeakTflops":2250.0,"mfuNominalPct":7.473474842848111,"configuredPeakTflops":2250.0,"mfuConfiguredPct":7.473474842848111,"vramAllocatedGb":26.684465664,"vramAllocatedPct":13.934227967673293,"vramReservedGb":27.145535488,"vramReservedPct":14.174991718296132,"warmupSteps":19,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":44859.07621225501,"meanTps":44327.48594263058,"stepMs":182.6163330078125,"jitter":0.011472550732189253,"achievedTflops":323.161948503978,"nominalPeakTflops":2250.0,"mfuNominalPct":14.362753266843468,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.362753266843468,"vramAllocatedGb":39.41270272,"vramAllocatedPct":20.580722561123764,"vramReservedGb":40.187723776,"vramReservedPct":20.985426938945363,"warmupSteps":16,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":56247.096226457696,"meanTps":56219.34073560202,"stepMs":291.28614807128906,"jitter":0.0016091370076547445,"achievedTflops":405.2005245990126,"nominalPeakTflops":2250.0,"mfuNominalPct":18.008912204400563,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.008912204400563,"vramAllocatedGb":64.869176832,"vramAllocatedPct":33.87371174802471,"vramReservedGb":66.339209216,"vramReservedPct":34.64134062305791,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":61406.98376721546,"meanTps":61404.8598752445,"stepMs":533.6200866699219,"jitter":0.000446331740466461,"achievedTflops":442.3720281726226,"nominalPeakTflops":2250.0,"mfuNominalPct":19.66097902989434,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.66097902989434,"vramAllocatedGb":115.782125056,"vramAllocatedPct":60.459690121826604,"vramReservedGb":118.277275648,"vramReservedPct":61.76262035848142,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.339817472,"vramAllocatedPct":99.39259947626779,"vramReservedGb":190.587076608,"vramReservedPct":99.52171449065467,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 19.81 MiB is free. Including non-PyTorch memory, this process has 178.32 GiB memory in use. Of the allocated memory 177.27 GiB is allocated by PyTorch, and 235.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":19144.060172017755,"meanTps":19134.576853346545,"stepMs":106.97835159301758,"jitter":0.0009951764410251658,"achievedTflops":137.9125989620031,"nominalPeakTflops":2250.0,"mfuNominalPct":6.129448842755694,"configuredPeakTflops":2250.0,"mfuConfiguredPct":6.129448842755694,"vramAllocatedGb":20.320330752,"vramAllocatedPct":7.069695872236231,"vramReservedGb":20.451426304,"vramReservedPct":7.1153056457263455,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":30470.677549300584,"meanTps":30300.233336854162,"stepMs":134.42431640625,"jitter":0.0007881164389159501,"achievedTflops":219.50883434327864,"nominalPeakTflops":2250.0,"mfuNominalPct":9.755948193034607,"configuredPeakTflops":2250.0,"mfuConfiguredPct":9.755948193034607,"vramAllocatedGb":26.684465664,"vramAllocatedPct":9.283857583816273,"vramReservedGb":27.145535488,"vramReservedPct":9.444269511718808,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":42184.72054840751,"meanTps":42162.57360345596,"stepMs":194.19353485107422,"jitter":0.0011990303431750691,"achievedTflops":303.896059406478,"nominalPeakTflops":2250.0,"mfuNominalPct":13.5064915291768,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.5064915291768,"vramAllocatedGb":39.41270272,"vramAllocatedPct":13.712169606581494,"vramReservedGb":40.187723776,"vramReservedPct":13.981809073939084,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":51951.5232834832,"meanTps":51936.638472834224,"stepMs":315.3709259033203,"jitter":0.0005054346956759407,"achievedTflops":374.2554887354927,"nominalPeakTflops":2250.0,"mfuNominalPct":16.633577277133007,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.633577277133007,"vramAllocatedGb":64.869176832,"vramAllocatedPct":22.568793652111935,"vramReservedGb":66.339209216,"vramReservedPct":23.08023620706127,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":59418.88326843921,"meanTps":59404.23792700707,"stepMs":551.4745178222656,"jitter":0.0002914214859321852,"achievedTflops":428.04987789100926,"nominalPeakTflops":2250.0,"mfuNominalPct":19.02443901737819,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.02443901737819,"vramAllocatedGb":115.782125056,"vramAllocatedPct":40.28204174317282,"vramReservedGb":118.277275648,"vramReservedPct":41.150135676099275,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":64162.16254815149,"meanTps":64126.36078704126,"stepMs":1021.4119567871094,"jitter":0.00047853917822269126,"achievedTflops":462.22016189502114,"nominalPeakTflops":2250.0,"mfuNominalPct":20.543118306445386,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.543118306445386,"vramAllocatedGb":217.608021504,"vramAllocatedPct":75.70853792529458,"vramReservedGb":222.94822912,"vramReservedPct":77.56646259199833,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.994584064,"vramAllocatedPct":99.50107383169323,"vramReservedGb":286.447894528,"vramReservedPct":99.65878618171818,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 679.62 MiB is free. Including non-PyTorch memory, this process has 267.01 GiB memory in use. Of the allocated memory 265.92 GiB is allocated by PyTorch, and 280.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":18315.086291109386,"meanTps":18324.94666456995,"stepMs":111.82038497924805,"jitter":0.001729571017243075,"achievedTflops":131.94072354161565,"nominalPeakTflops":2250.0,"mfuNominalPct":5.86403215740514,"configuredPeakTflops":2250.0,"mfuConfiguredPct":5.86403215740514,"vramAllocatedGb":20.320330752,"vramAllocatedPct":7.069695872236231,"vramReservedGb":20.451426304,"vramReservedPct":7.1153056457263455,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":30320.759052908663,"meanTps":30093.524013780832,"stepMs":135.0889663696289,"jitter":0.0004205047434594342,"achievedTflops":218.4288309749176,"nominalPeakTflops":2250.0,"mfuNominalPct":9.70794804332967,"configuredPeakTflops":2250.0,"mfuConfiguredPct":9.70794804332967,"vramAllocatedGb":26.684465664,"vramAllocatedPct":9.283857583816273,"vramReservedGb":27.145535488,"vramReservedPct":9.444269511718808,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":42119.02182862631,"meanTps":42109.344110489736,"stepMs":194.49644470214844,"jitter":0.0005221340351824682,"achievedTflops":303.42276998343567,"nominalPeakTflops":2250.0,"mfuNominalPct":13.485456443708253,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.485456443708253,"vramAllocatedGb":39.41270272,"vramAllocatedPct":13.712169606581494,"vramReservedGb":40.187723776,"vramReservedPct":13.981809073939084,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":51839.46276382263,"meanTps":51825.305952631184,"stepMs":316.0526580810547,"jitter":0.0005211654884193258,"achievedTflops":373.44821183766874,"nominalPeakTflops":2250.0,"mfuNominalPct":16.597698303896387,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.597698303896387,"vramAllocatedGb":64.869176832,"vramAllocatedPct":22.568793652111935,"vramReservedGb":66.339209216,"vramReservedPct":23.08023620706127,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":59363.88105350337,"meanTps":59365.81194414068,"stepMs":551.9854736328125,"jitter":0.0002782826886693875,"achievedTflops":427.6536454125115,"nominalPeakTflops":2250.0,"mfuNominalPct":19.00682868500051,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.00682868500051,"vramAllocatedGb":115.782125056,"vramAllocatedPct":40.28204174317282,"vramReservedGb":118.277275648,"vramReservedPct":41.150135676099275,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":32,"tokensPerStep":65536,"status":"complete","stable":true,"tps":64145.19176825281,"meanTps":64134.82494202773,"stepMs":1021.6821899414062,"jitter":0.00032974731650019256,"achievedTflops":462.0979054697273,"nominalPeakTflops":2250.0,"mfuNominalPct":20.537684687543436,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.537684687543436,"vramAllocatedGb":217.608021504,"vramAllocatedPct":75.70853792529458,"vramReservedGb":222.94822912,"vramReservedPct":77.56646259199833,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.994584064,"vramAllocatedPct":99.50107383169323,"vramReservedGb":286.447894528,"vramReservedPct":99.65878618171818,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 679.62 MiB is free. Including non-PyTorch memory, this process has 267.01 GiB memory in use. Of the allocated memory 265.92 GiB is allocated by PyTorch, and 280.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":10972.758900525989,"meanTps":10973.635174456424,"stepMs":186.64403533935547,"jitter":0.0009499647785808643,"achievedTflops":79.04706128989842,"nominalPeakTflops":165.2,"mfuNominalPct":47.84931070817096,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":38.46260382283599,"vramAllocatedGb":20.320330752,"vramAllocatedPct":80.44136136571774,"vramReservedGb":20.451426304,"vramReservedPct":80.96032460507507,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.586116096,"vramAllocatedPct":97.32817214420434,"vramReservedGb":24.750587904,"vramReservedPct":97.97926076590402,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 14.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 9.69 MiB is free. Including non-PyTorch memory, this process has 23.51 GiB memory in use. Of the allocated memory 22.90 GiB is allocated by PyTorch, and 154.85 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":10970.854771321608,"meanTps":10969.045384850286,"stepMs":186.67642974853516,"jitter":0.001002637283808401,"achievedTflops":79.03334406351198,"nominalPeakTflops":165.2,"mfuNominalPct":47.84100730236803,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":38.45592931482215,"vramAllocatedGb":20.320330752,"vramAllocatedPct":80.44136136571774,"vramReservedGb":20.451426304,"vramReservedPct":80.96032460507507,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.586116096,"vramAllocatedPct":97.32817214420434,"vramReservedGb":24.750587904,"vramReservedPct":97.97926076590402,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 14.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 9.69 MiB is free. Including non-PyTorch memory, this process has 23.51 GiB memory in use. Of the allocated memory 22.90 GiB is allocated by PyTorch, and 154.85 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":14511.442848036177,"meanTps":14508.34756209397,"stepMs":141.1300048828125,"jitter":0.0001951612892762709,"achievedTflops":104.53951668969846,"nominalPeakTflops":209.5,"mfuNominalPct":49.89953063947421,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":38.8699580094545,"vramAllocatedGb":20.320330752,"vramAllocatedPct":60.35348997851078,"vramReservedGb":20.451426304,"vramReservedPct":60.74285736531282,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":17195.751455176025,"meanTps":17193.584657785057,"stepMs":238.19837188720703,"jitter":0.00036503302874587814,"achievedTflops":123.87710616133207,"nominalPeakTflops":209.5,"mfuNominalPct":59.12988360922772,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":46.06007438427544,"vramAllocatedGb":26.684465664,"vramAllocatedPct":79.25563076160203,"vramReservedGb":27.145535488,"vramReservedPct":80.62505596150626,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.84903936,"vramAllocatedPct":97.56505403448396,"vramReservedGb":33.017561088,"vramReservedPct":98.06558104588649,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 9.88 MiB is free. Including non-PyTorch memory, this process has 31.34 GiB memory in use. Of the allocated memory 30.59 GiB is allocated by PyTorch, and 160.71 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":14502.30706956217,"meanTps":14502.932714551232,"stepMs":141.21891021728516,"jitter":0.00042411846236421746,"achievedTflops":104.47370311924524,"nominalPeakTflops":209.5,"mfuNominalPct":49.86811604737243,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":38.84548716052603,"vramAllocatedGb":20.320330752,"vramAllocatedPct":60.35348997851078,"vramReservedGb":20.451426304,"vramReservedPct":60.74285736531282,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":17193.430261410664,"meanTps":17192.52100891628,"stepMs":238.23052978515625,"jitter":0.0002880651017160671,"achievedTflops":123.86038442822014,"nominalPeakTflops":209.5,"mfuNominalPct":59.12190187504542,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":46.05385689748683,"vramAllocatedGb":26.684465664,"vramAllocatedPct":79.25563076160203,"vramReservedGb":27.145535488,"vramReservedPct":80.62505596150626,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":null,"error":null},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.84903936,"vramAllocatedPct":97.56505403448396,"vramReservedGb":33.017561088,"vramReservedPct":98.06558104588649,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 9.88 MiB is free. Including non-PyTorch memory, this process has 31.34 GiB memory in use. Of the allocated memory 30.59 GiB is allocated by PyTorch, and 160.71 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":23982.3879595725,"meanTps":23976.125648314908,"stepMs":85.39599990844727,"jitter":0.0013178196295380216,"achievedTflops":172.76760640640484,"nominalPeakTflops":989.5,"mfuNominalPct":17.46009160246638,"configuredPeakTflops":989.5,"mfuConfiguredPct":17.46009160246638,"vramAllocatedGb":20.370400256,"vramAllocatedPct":23.96024561114777,"vramReservedGb":20.493369344,"vramReservedPct":24.104885358724207,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":30411.60201184015,"meanTps":30409.750742820637,"stepMs":134.68544006347656,"jitter":0.0007413214157653812,"achievedTflops":219.08325790687803,"nominalPeakTflops":989.5,"mfuNominalPct":22.14080423515695,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.14080423515695,"vramAllocatedGb":26.734535168,"vramAllocatedPct":31.445922557975862,"vramReservedGb":27.187478528,"vramReservedPct":31.978687452977958,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":35304.09175484126,"meanTps":35303.138083097365,"stepMs":232.0410919189453,"jitter":0.00038714981115660536,"achievedTflops":254.3284446535466,"nominalPeakTflops":989.5,"mfuNominalPct":25.70272305745797,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.70272305745797,"vramAllocatedGb":39.462772224,"vramAllocatedPct":46.41723790897612,"vramReservedGb":40.229666816,"vramReservedPct":47.319281194884,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":39075.064580337705,"meanTps":39075.20080839464,"stepMs":419.2955322265625,"jitter":0.0004523505263228379,"achievedTflops":281.49429444226934,"nominalPeakTflops":989.5,"mfuNominalPct":28.44813486025966,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.44813486025966,"vramAllocatedGb":64.919246336,"vramAllocatedPct":76.35986861097663,"vramReservedGb":66.360180736,"vramReservedPct":78.05473673824295,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.98025728,"vramAllocatedPct":98.77997317816573,"vramReservedGb":84.16919552,"vramReservedPct":99.00220772333157,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 56.19 MiB is free. Including non-PyTorch memory, this process has 79.12 GiB memory in use. Of the allocated memory 78.21 GiB is allocated by PyTorch, and 180.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":23935.859299990883,"meanTps":23936.491820447787,"stepMs":85.5620002746582,"jitter":0.0010412773408668425,"achievedTflops":172.43241688487907,"nominalPeakTflops":989.5,"mfuNominalPct":17.4262169666376,"configuredPeakTflops":989.5,"mfuConfiguredPct":17.4262169666376,"vramAllocatedGb":20.370400256,"vramAllocatedPct":23.96024561114777,"vramReservedGb":20.493369344,"vramReservedPct":24.104885358724207,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":30348.7692706613,"meanTps":30343.365110381226,"stepMs":134.96428680419922,"jitter":0.0010243243217270463,"achievedTflops":218.6306148124655,"nominalPeakTflops":989.5,"mfuNominalPct":22.095059607121325,"configuredPeakTflops":989.5,"mfuConfiguredPct":22.095059607121325,"vramAllocatedGb":26.734535168,"vramAllocatedPct":31.445922557975862,"vramReservedGb":27.187478528,"vramReservedPct":31.978687452977958,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":35289.796725936445,"meanTps":35281.27627222376,"stepMs":232.1350860595703,"jitter":0.0006024656342959948,"achievedTflops":254.2254641692196,"nominalPeakTflops":989.5,"mfuNominalPct":25.692315732109105,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.692315732109105,"vramAllocatedGb":39.462772224,"vramAllocatedPct":46.41723790897612,"vramReservedGb":40.229666816,"vramReservedPct":47.319281194884,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":38959.218866339965,"meanTps":38913.58601542306,"stepMs":420.5423126220703,"jitter":0.0009200399003492694,"achievedTflops":280.6597492437857,"nominalPeakTflops":989.5,"mfuNominalPct":28.36379476945788,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.36379476945788,"vramAllocatedGb":64.919246336,"vramAllocatedPct":76.35986861097663,"vramReservedGb":66.360180736,"vramReservedPct":78.05473673824295,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.98025728,"vramAllocatedPct":98.77997317816573,"vramReservedGb":84.16919552,"vramReservedPct":99.00220772333157,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 56.19 MiB is free. Including non-PyTorch memory, this process has 79.12 GiB memory in use. Of the allocated memory 78.21 GiB is allocated by PyTorch, and 180.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":25501.574946511006,"meanTps":25496.9251174671,"stepMs":80.30876541137695,"jitter":0.000808961299108926,"achievedTflops":183.71173339907824,"nominalPeakTflops":989.5,"mfuNominalPct":18.56611757443944,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.56611757443944,"vramAllocatedGb":20.370400256,"vramAllocatedPct":13.569342675542126,"vramReservedGb":20.493369344,"vramReservedPct":13.651256122141163,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":32894.66876830608,"meanTps":32899.00557107653,"stepMs":124.51865768432617,"jitter":0.0005879198827736898,"achievedTflops":236.97111381118143,"nominalPeakTflops":989.5,"mfuNominalPct":23.948571380614595,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.948571380614595,"vramAllocatedGb":26.734535168,"vramAllocatedPct":17.808686349158606,"vramReservedGb":27.187478528,"vramReservedPct":18.110405686393577,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":38561.157222047084,"meanTps":38556.495576359375,"stepMs":212.44175720214844,"jitter":0.0014500055700871024,"achievedTflops":277.7921383285337,"nominalPeakTflops":989.5,"mfuNominalPct":28.07399073557693,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.07399073557693,"vramAllocatedGb":39.462772224,"vramAllocatedPct":26.28735186863093,"vramReservedGb":40.229666816,"vramReservedPct":26.798203662610923,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":42137.21850799244,"meanTps":42078.00355272335,"stepMs":388.8249053955078,"jitter":0.003115015739131291,"achievedTflops":303.5538577109768,"nominalPeakTflops":989.5,"mfuNominalPct":30.677499516015843,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.677499516015843,"vramAllocatedGb":64.919246336,"vramAllocatedPct":43.24468290757559,"vramReservedGb":66.360180736,"vramReservedPct":44.20453310201728,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":43863.864320058354,"meanTps":43871.005131012054,"stepMs":747.0386047363281,"jitter":0.0016366224530526437,"achievedTflops":315.9925049618317,"nominalPeakTflops":989.5,"mfuNominalPct":31.934563412009265,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.934563412009265,"vramAllocatedGb":115.83219456,"vramAllocatedPct":77.1593449854649,"vramReservedGb":118.340190208,"vramReservedPct":78.82999710563894,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.081375744,"vramAllocatedPct":99.30763502871005,"vramReservedGb":149.32770816,"vramReservedPct":99.47172453715325,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 45.44 MiB is free. Including non-PyTorch memory, this process has 139.76 GiB memory in use. Of the allocated memory 138.84 GiB is allocated by PyTorch, and 194.92 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":25668.766991418786,"meanTps":25668.88750990498,"stepMs":79.78567886352539,"jitter":0.0011173725673174666,"achievedTflops":184.91617431870654,"nominalPeakTflops":989.5,"mfuNominalPct":18.68783974923765,"configuredPeakTflops":989.5,"mfuConfiguredPct":18.68783974923765,"vramAllocatedGb":20.370400256,"vramAllocatedPct":13.569342675542126,"vramReservedGb":20.493369344,"vramReservedPct":13.651256122141163,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":33080.0975154786,"meanTps":33075.20290447682,"stepMs":123.82067489624023,"jitter":0.0009094607889183494,"achievedTflops":238.30693078078173,"nominalPeakTflops":989.5,"mfuNominalPct":24.083570569053233,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.083570569053233,"vramAllocatedGb":26.734535168,"vramAllocatedPct":17.808686349158606,"vramReservedGb":27.187478528,"vramReservedPct":18.110405686393577,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":38696.83506312056,"meanTps":38701.429993314356,"stepMs":211.6968994140625,"jitter":0.001136333482007195,"achievedTflops":278.76955291644583,"nominalPeakTflops":989.5,"mfuNominalPct":28.172769370029897,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.172769370029897,"vramAllocatedGb":39.462772224,"vramAllocatedPct":26.28735186863093,"vramReservedGb":40.229666816,"vramReservedPct":26.798203662610923,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":42112.98894091743,"meanTps":42117.87078807249,"stepMs":389.0486145019531,"jitter":0.003994642920316324,"achievedTflops":303.37930944185246,"nominalPeakTflops":989.5,"mfuNominalPct":30.659859468605603,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.659859468605603,"vramAllocatedGb":64.919246336,"vramAllocatedPct":43.24468290757559,"vramReservedGb":66.360180736,"vramReservedPct":44.20453310201728,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":43941.29259468075,"meanTps":43953.00778381664,"stepMs":745.7222595214844,"jitter":0.0021928510624993373,"achievedTflops":316.5502933562668,"nominalPeakTflops":989.5,"mfuNominalPct":31.99093414414015,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.99093414414015,"vramAllocatedGb":115.83219456,"vramAllocatedPct":77.1593449854649,"vramReservedGb":118.340190208,"vramReservedPct":78.82999710563894,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.081375744,"vramAllocatedPct":99.30763502871005,"vramReservedGb":149.32770816,"vramReservedPct":99.47172453715325,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 45.44 MiB is free. Including non-PyTorch memory, this process has 139.76 GiB memory in use. Of the allocated memory 138.84 GiB is allocated by PyTorch, and 194.92 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":9444.708518110916,"meanTps":9438.968411168453,"stepMs":216.84099578857422,"jitter":0.006256054308767328,"achievedTflops":68.03908295666201,"nominalPeakTflops":364.2,"mfuNominalPct":18.681791036974744,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.071783107124551,"vramAllocatedGb":20.320330752,"vramAllocatedPct":39.94023598868506,"vramReservedGb":20.451426304,"vramReservedPct":40.19790833407401,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":10868.688080903537,"meanTps":10877.82555734482,"stepMs":376.86241149902344,"jitter":0.00262224761423521,"achievedTflops":78.29734168594435,"nominalPeakTflops":364.2,"mfuNominalPct":21.498446371758472,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.34415721779548,"vramAllocatedGb":26.684465664,"vramAllocatedPct":52.449139182797275,"vramReservedGb":27.145535488,"vramReservedPct":53.3553861234879,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10999.144646732135,"meanTps":10999.24181636041,"stepMs":744.7851867675781,"jitter":0.0015165516450420385,"achievedTflops":79.237142537144,"nominalPeakTflops":364.2,"mfuNominalPct":21.756491635679296,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.55233866260117,"vramAllocatedGb":39.41270272,"vramAllocatedPct":77.46688116450841,"vramReservedGb":40.187723776,"vramReservedPct":78.99020892184785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":49.861430272,"vramAllocatedPct":98.00417700391205,"vramReservedGb":50.180653056,"vramReservedPct":98.63161921838832,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 152.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 138.50 MiB is free. Process 2308332 has 47.24 GiB memory in use. Of the allocated memory 46.44 GiB is allocated by PyTorch, and 304.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":9501.303584353476,"meanTps":9498.2832952912,"stepMs":215.54936981201172,"jitter":0.001960847963579277,"achievedTflops":68.4467902352542,"nominalPeakTflops":364.2,"mfuNominalPct":18.793737022310324,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.162097017999152,"vramAllocatedGb":20.320330752,"vramAllocatedPct":39.94023598868506,"vramReservedGb":20.451426304,"vramReservedPct":40.19790833407401,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":10903.065017835164,"meanTps":10898.449191498723,"stepMs":375.67417907714844,"jitter":0.004614513035910415,"achievedTflops":78.54499096587732,"nominalPeakTflops":364.2,"mfuNominalPct":21.566444526600034,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.39901563257104,"vramAllocatedGb":26.684465664,"vramAllocatedPct":52.449139182797275,"vramReservedGb":27.145535488,"vramReservedPct":53.3553861234879,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10979.606847015748,"meanTps":10978.707617290349,"stepMs":746.1105041503906,"jitter":0.00160948965110363,"achievedTflops":79.0963934634013,"nominalPeakTflops":364.2,"mfuNominalPct":21.7178455418455,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.521160412986465,"vramAllocatedGb":39.41270272,"vramAllocatedPct":77.46688116450841,"vramReservedGb":40.187723776,"vramReservedPct":78.99020892184785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":49.861430272,"vramAllocatedPct":98.00417700391205,"vramReservedGb":50.180653056,"vramReservedPct":98.63161921838832,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 152.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 138.50 MiB is free. Process 2312978 has 47.24 GiB memory in use. Of the allocated memory 46.44 GiB is allocated by PyTorch, and 304.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":6140.059778533234,"meanTps":6140.73261040287,"stepMs":333.5472412109375,"jitter":0.0029101465679311996,"achievedTflops":44.2326023962935,"nominalPeakTflops":154.8,"mfuNominalPct":28.574032555745156,"configuredPeakTflops":180.6,"mfuConfiguredPct":24.492027904924424,"vramAllocatedGb":20.320330752,"vramAllocatedPct":39.817294190051314,"vramReservedGb":20.451426304,"vramReservedPct":40.074173382850745,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7476.91420488451,"meanTps":7474.730200387086,"stepMs":547.819580078125,"jitter":0.0013287355901792839,"achievedTflops":53.863217151749,"nominalPeakTflops":154.8,"mfuNominalPct":34.79535991715051,"configuredPeakTflops":180.6,"mfuConfiguredPct":29.82459421470044,"vramAllocatedGb":26.684465664,"vramAllocatedPct":52.287693178578586,"vramReservedGb":27.145535488,"vramReservedPct":53.191150560666536,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":8478.065366472505,"meanTps":8478.892090872823,"stepMs":966.2581787109375,"jitter":0.0005500103656241791,"achievedTflops":61.075446815038674,"nominalPeakTflops":154.8,"mfuNominalPct":39.45442300713093,"configuredPeakTflops":180.6,"mfuConfiguredPct":33.81807686325508,"vramAllocatedGb":39.41270272,"vramAllocatedPct":77.22842694737234,"vramReservedGb":40.187723776,"vramReservedPct":78.74706568248244,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.338335744,"vramAllocatedPct":98.6369930597291,"vramReservedGb":50.642026496,"vramReservedPct":99.23206920109513,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 36.25 MiB is free. Process 1348507 has 47.49 GiB memory in use. Of the allocated memory 46.88 GiB is allocated by PyTorch, and 289.62 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":6152.622646783699,"meanTps":6153.418553739326,"stepMs":332.8661804199219,"jitter":0.0016564692607990625,"achievedTflops":44.323104504794564,"nominalPeakTflops":154.8,"mfuNominalPct":28.632496450125686,"configuredPeakTflops":180.6,"mfuConfiguredPct":24.542139814393447,"vramAllocatedGb":20.320330752,"vramAllocatedPct":39.817294190051314,"vramReservedGb":20.451426304,"vramReservedPct":40.074173382850745,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7471.410626983984,"meanTps":7470.55637028164,"stepMs":548.2231140136719,"jitter":0.0009085829658194712,"achievedTflops":53.82356972455585,"nominalPeakTflops":154.8,"mfuNominalPct":34.769747884080004,"configuredPeakTflops":180.6,"mfuConfiguredPct":29.802641043497154,"vramAllocatedGb":26.684465664,"vramAllocatedPct":52.287693178578586,"vramReservedGb":27.145535488,"vramReservedPct":53.191150560666536,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":8474.833070366174,"meanTps":8472.83839285198,"stepMs":966.626708984375,"jitter":0.0007138608929399947,"achievedTflops":61.052161558273205,"nominalPeakTflops":154.8,"mfuNominalPct":39.43938085159767,"configuredPeakTflops":180.6,"mfuConfiguredPct":33.805183587083725,"vramAllocatedGb":39.41270272,"vramAllocatedPct":77.22842694737234,"vramReservedGb":40.187723776,"vramReservedPct":78.74706568248244,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.338335744,"vramAllocatedPct":98.6369930597291,"vramReservedGb":50.642026496,"vramReservedPct":99.23206920109513,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 36.25 MiB is free. Process 1380474 has 47.49 GiB memory in use. Of the allocated memory 46.88 GiB is allocated by PyTorch, and 289.62 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":16728.645423553055,"meanTps":16722.44596562402,"stepMs":122.42473602294922,"jitter":0.0004524545888795999,"achievedTflops":120.51210384556876,"nominalPeakTflops":468.0,"mfuNominalPct":25.750449539651445,"configuredPeakTflops":468.0,"mfuConfiguredPct":25.750449539651445,"vramAllocatedGb":20.320330752,"vramAllocatedPct":19.92698199838175,"vramReservedGb":20.451426304,"vramReservedPct":20.055539881452365,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":21764.159642051185,"meanTps":21764.169964910274,"stepMs":188.1993179321289,"jitter":0.00026559508013345893,"achievedTflops":156.78762986999425,"nominalPeakTflops":468.0,"mfuNominalPct":33.501630314101334,"configuredPeakTflops":468.0,"mfuConfiguredPct":33.501630314101334,"vramAllocatedGb":26.684465664,"vramAllocatedPct":26.167923810523014,"vramReservedGb":27.145535488,"vramReservedPct":26.620068521894936,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":24584.26853098999,"meanTps":24582.38505846282,"stepMs":333.2212219238281,"jitter":0.00026246612668402665,"achievedTflops":177.10351598478417,"nominalPeakTflops":468.0,"mfuNominalPct":37.84263162068038,"configuredPeakTflops":468.0,"mfuConfiguredPct":37.84263162068038,"vramAllocatedGb":39.41270272,"vramAllocatedPct":38.649775301108804,"vramReservedGb":40.187723776,"vramReservedPct":39.40979396516322,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":24358.54195633419,"meanTps":24357.192607279987,"stepMs":672.6182556152344,"jitter":0.0002738460700837169,"achievedTflops":175.4773960141065,"nominalPeakTflops":468.0,"mfuNominalPct":37.495170088484294,"configuredPeakTflops":468.0,"mfuConfiguredPct":37.495170088484294,"vramAllocatedGb":64.869176832,"vramAllocatedPct":63.61347828228039,"vramReservedGb":66.339209216,"vramReservedPct":65.05505466263153,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":101.057241088,"vramAllocatedPct":99.10103573331345,"vramReservedGb":101.206458368,"vramReservedPct":99.24736455486153,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 61.75 MiB is free. Including non-PyTorch memory, this process has 94.90 GiB memory in use. Of the allocated memory 94.12 GiB is allocated by PyTorch, and 142.30 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":16721.669227586775,"meanTps":16711.85639964196,"stepMs":122.47581100463867,"jitter":0.00034979724695144714,"achievedTflops":120.46184777094652,"nominalPeakTflops":468.0,"mfuNominalPct":25.73971106216806,"configuredPeakTflops":468.0,"mfuConfiguredPct":25.73971106216806,"vramAllocatedGb":20.320330752,"vramAllocatedPct":19.92698199838175,"vramReservedGb":20.451426304,"vramReservedPct":20.055539881452365,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":21760.61162912964,"meanTps":21758.935414635107,"stepMs":188.2300033569336,"jitter":0.0006858435154180604,"achievedTflops":156.76207020925537,"nominalPeakTflops":468.0,"mfuNominalPct":33.49616884813149,"configuredPeakTflops":468.0,"mfuConfiguredPct":33.49616884813149,"vramAllocatedGb":26.684465664,"vramAllocatedPct":26.167923810523014,"vramReservedGb":27.145535488,"vramReservedPct":26.620068521894936,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":24591.20740438362,"meanTps":24592.532676825092,"stepMs":333.127197265625,"jitter":0.00012950444268880143,"achievedTflops":177.153503189953,"nominalPeakTflops":468.0,"mfuNominalPct":37.85331264742586,"configuredPeakTflops":468.0,"mfuConfiguredPct":37.85331264742586,"vramAllocatedGb":39.41270272,"vramAllocatedPct":38.649775301108804,"vramReservedGb":40.187723776,"vramReservedPct":39.40979396516322,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":24352.803007990413,"meanTps":24352.47382420825,"stepMs":672.7767639160156,"jitter":0.00017545960864934687,"achievedTflops":175.43605299312313,"nominalPeakTflops":468.0,"mfuNominalPct":37.48633610964169,"configuredPeakTflops":468.0,"mfuConfiguredPct":37.48633610964169,"vramAllocatedGb":64.869176832,"vramAllocatedPct":63.61347828228039,"vramReservedGb":66.339209216,"vramReservedPct":65.05505466263153,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":101.057241088,"vramAllocatedPct":99.10103573331345,"vramReservedGb":101.206458368,"vramReservedPct":99.24736455486153,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 61.75 MiB is free. Including non-PyTorch memory, this process has 94.90 GiB memory in use. Of the allocated memory 94.12 GiB is allocated by PyTorch, and 142.30 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14782.344194351357,"meanTps":14779.790009567163,"stepMs":554.1746215820312,"jitter":0.001887867919688506,"achievedTflops":124.71955660174001,"nominalPeakTflops":312.0,"mfuNominalPct":39.974216859532056,"configuredPeakTflops":312.0,"mfuConfiguredPct":39.974216859532056,"vramAllocatedGb":39.420682752,"vramAllocatedPct":46.39133326848444,"vramReservedGb":40.08706048,"vramReservedPct":47.175544730696494,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":15757.01062072426,"meanTps":15759.457249288705,"stepMs":1039.7911376953125,"jitter":0.0005998825424229649,"achievedTflops":132.9428778107186,"nominalPeakTflops":312.0,"mfuNominalPct":42.60989673420469,"configuredPeakTflops":312.0,"mfuConfiguredPct":42.60989673420469,"vramAllocatedGb":64.877796864,"vramAllocatedPct":76.34995859857752,"vramReservedGb":66.204991488,"vramReservedPct":77.91183738442886,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.329465856,"vramAllocatedPct":99.24121252518117,"vramReservedGb":84.408270848,"vramReservedPct":99.33395238638782,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 14.75 MiB is free. Process 764857 has 79.12 GiB memory in use. Of the allocated memory 78.54 GiB is allocated by PyTorch, and 75.15 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14746.360179187459,"meanTps":14744.568330185404,"stepMs":555.5269165039062,"jitter":0.0014195197013146808,"achievedTflops":124.41595722961156,"nominalPeakTflops":312.0,"mfuNominalPct":39.87690936846524,"configuredPeakTflops":312.0,"mfuConfiguredPct":39.87690936846524,"vramAllocatedGb":39.420682752,"vramAllocatedPct":46.39133326848444,"vramReservedGb":40.08706048,"vramReservedPct":47.175544730696494,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":15750.231044263744,"meanTps":15749.84818570899,"stepMs":1040.2387084960938,"jitter":0.0006702605972761918,"achievedTflops":132.88567810281123,"nominalPeakTflops":312.0,"mfuNominalPct":42.59156349449078,"configuredPeakTflops":312.0,"mfuConfiguredPct":42.59156349449078,"vramAllocatedGb":64.877796864,"vramAllocatedPct":76.34995859857752,"vramReservedGb":66.204991488,"vramReservedPct":77.91183738442886,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.329465856,"vramAllocatedPct":99.24121252518117,"vramReservedGb":84.408270848,"vramReservedPct":99.33395238638782,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 14.75 MiB is free. Process 852916 has 79.12 GiB memory in use. Of the allocated memory 78.54 GiB is allocated by PyTorch, and 75.15 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":49591.59418964524,"meanTps":49524.30357690651,"stepMs":330.3785705566406,"jitter":0.0025290565493963748,"achievedTflops":418.4073619980663,"nominalPeakTflops":2250.0,"mfuNominalPct":18.59588275546961,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.59588275546961,"vramAllocatedGb":64.877796864,"vramAllocatedPct":33.87821299952021,"vramReservedGb":66.169339904,"vramReservedPct":34.55263741468476,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":53602.77403644068,"meanTps":53561.24290026289,"stepMs":611.3116455078125,"jitter":0.0012705910702465024,"achievedTflops":452.24993563623974,"nominalPeakTflops":2250.0,"mfuNominalPct":20.099997139388435,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.099997139388435,"vramAllocatedGb":115.792025088,"vramAllocatedPct":60.46485977013481,"vramReservedGb":118.28985856,"vramReservedPct":61.769190966509065,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.347157504,"vramAllocatedPct":99.39643233095057,"vramReservedGb":190.59965952,"vramReservedPct":99.52828509868232,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 7.81 MiB is free. Including non-PyTorch memory, this process has 178.33 GiB memory in use. Of the allocated memory 177.27 GiB is allocated by PyTorch, and 240.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":49668.423982424436,"meanTps":49328.43669930408,"stepMs":329.8675231933594,"jitter":0.0008469081187988008,"achievedTflops":419.05557973425493,"nominalPeakTflops":2250.0,"mfuNominalPct":18.62469243263355,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.62469243263355,"vramAllocatedGb":64.877796864,"vramAllocatedPct":33.87821299952021,"vramReservedGb":66.169339904,"vramReservedPct":34.55263741468476,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":53634.82138481942,"meanTps":53620.98189142606,"stepMs":610.9463806152344,"jitter":0.0006501536627479689,"achievedTflops":452.52032110606154,"nominalPeakTflops":2250.0,"mfuNominalPct":20.112014271380513,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.112014271380513,"vramAllocatedGb":115.792025088,"vramAllocatedPct":60.46485977013481,"vramReservedGb":118.28985856,"vramReservedPct":61.769190966509065,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.347157504,"vramAllocatedPct":99.39643233095057,"vramReservedGb":190.59965952,"vramReservedPct":99.52828509868232,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 7.81 MiB is free. Including non-PyTorch memory, this process has 178.33 GiB memory in use. Of the allocated memory 177.27 GiB is allocated by PyTorch, and 240.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":52358.16760475663,"meanTps":52333.16293347024,"stepMs":625.8431396484375,"jitter":0.0002674916865080642,"achievedTflops":441.7491138272993,"nominalPeakTflops":2250.0,"mfuNominalPct":19.63329394787997,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.63329394787997,"vramAllocatedGb":115.792025088,"vramAllocatedPct":40.28548608747113,"vramReservedGb":118.252109824,"vramReservedPct":41.14138017284367,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":56090.765605732115,"meanTps":56085.431567767395,"stepMs":1168.3919677734375,"jitter":0.0002432671292707795,"achievedTflops":473.2412751200997,"nominalPeakTflops":2250.0,"mfuNominalPct":21.03294556089332,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.03294556089332,"vramAllocatedGb":217.620481536,"vramAllocatedPct":75.71287292544164,"vramReservedGb":222.971297792,"vramReservedPct":77.57448846998264,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.001924096,"vramAllocatedPct":99.50362752014279,"vramReservedGb":286.284316672,"vramReservedPct":99.60187541055672,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 757.62 MiB is free. Including non-PyTorch memory, this process has 266.93 GiB memory in use. Of the allocated memory 265.92 GiB is allocated by PyTorch, and 197.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":52383.60890614483,"meanTps":52378.99749674535,"stepMs":625.5391845703125,"jitter":0.0002499007691996963,"achievedTflops":441.9637636681358,"nominalPeakTflops":2250.0,"mfuNominalPct":19.642833940806035,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.642833940806035,"vramAllocatedGb":115.792025088,"vramAllocatedPct":40.28548608747113,"vramReservedGb":118.252109824,"vramReservedPct":41.14138017284367,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":16,"tokensPerStep":65536,"status":"complete","stable":true,"tps":56102.04290842311,"meanTps":56094.88651211471,"stepMs":1168.1571044921875,"jitter":0.00022914003312383881,"achievedTflops":473.3364224237203,"nominalPeakTflops":2250.0,"mfuNominalPct":21.037174329943124,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.037174329943124,"vramAllocatedGb":217.620481536,"vramAllocatedPct":75.71287292544164,"vramReservedGb":222.971297792,"vramReservedPct":77.57448846998264,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.001924096,"vramAllocatedPct":99.50362752014279,"vramReservedGb":286.284316672,"vramReservedPct":99.60187541055672,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 757.62 MiB is free. Including non-PyTorch memory, this process has 266.93 GiB memory in use. Of the allocated memory 265.92 GiB is allocated by PyTorch, and 197.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.657845248,"vramAllocatedPct":97.6121237543877,"vramReservedGb":24.754782208,"vramReservedPct":97.99586460606092,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 14.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 5.69 MiB is free. Including non-PyTorch memory, this process has 23.51 GiB memory in use. Of the allocated memory 22.96 GiB is allocated by PyTorch, and 93.98 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.657845248,"vramAllocatedPct":97.6121237543877,"vramReservedGb":24.754782208,"vramReservedPct":97.99586460606092,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 14.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 5.69 MiB is free. Including non-PyTorch memory, this process has 23.51 GiB memory in use. Of the allocated memory 22.96 GiB is allocated by PyTorch, and 93.98 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":16171.04732316262,"meanTps":16167.164468555471,"stepMs":253.2921905517578,"jitter":0.00040740836748393364,"achievedTflops":136.43613119908787,"nominalPeakTflops":209.5,"mfuNominalPct":65.12464496376509,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":50.72978007371604,"vramAllocatedGb":26.692092928,"vramAllocatedPct":79.2782845267895,"vramReservedGb":26.81208832,"vramReservedPct":79.63468328707182,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.856379392,"vramAllocatedPct":97.58685469083944,"vramReservedGb":33.00917248,"vramReservedPct":98.04066601005167,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 19.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.60 GiB is allocated by PyTorch, and 145.71 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":16171.00835630701,"meanTps":16171.639106655315,"stepMs":253.2928009033203,"jitter":0.0003725301339528429,"achievedTflops":136.43580243330553,"nominalPeakTflops":209.5,"mfuNominalPct":65.1244880349907,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":50.72965783178726,"vramAllocatedGb":26.692092928,"vramAllocatedPct":79.2782845267895,"vramReservedGb":26.81208832,"vramReservedPct":79.63468328707182,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.856379392,"vramAllocatedPct":97.58685469083944,"vramReservedGb":33.00917248,"vramReservedPct":98.04066601005167,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 19.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.60 GiB is allocated by PyTorch, and 145.71 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":32276.537317251783,"meanTps":32273.632617691426,"stepMs":253.80665588378906,"jitter":0.00045217434747391597,"achievedTflops":272.3191511387883,"nominalPeakTflops":989.5,"mfuNominalPct":27.520884400079666,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.520884400079666,"vramAllocatedGb":39.470752256,"vramAllocatedPct":46.42662425015263,"vramReservedGb":40.12900352,"vramReservedPct":47.200878155872665,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":35555.751942906914,"meanTps":35552.99506555107,"stepMs":460.7974548339844,"jitter":0.0003094434544499782,"achievedTflops":299.98608871895436,"nominalPeakTflops":989.5,"mfuNominalPct":30.316936707322323,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.316936707322323,"vramAllocatedGb":64.927866368,"vramAllocatedPct":76.37000773840174,"vramReservedGb":66.225963008,"vramReservedPct":77.89686601956117,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.105955328,"vramAllocatedPct":98.9278228063062,"vramReservedGb":84.198555648,"vramReservedPct":99.03674194304321,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 30.19 MiB is free. Including non-PyTorch memory, this process has 79.14 GiB memory in use. Of the allocated memory 78.33 GiB is allocated by PyTorch, and 88.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":32286.376491694577,"meanTps":32258.004085908775,"stepMs":253.72930908203125,"jitter":0.001154408053833669,"achievedTflops":272.4021648650078,"nominalPeakTflops":989.5,"mfuNominalPct":27.52927386205233,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.52927386205233,"vramAllocatedGb":39.470752256,"vramAllocatedPct":46.42662425015263,"vramReservedGb":40.12900352,"vramReservedPct":47.200878155872665,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":35567.28585076387,"meanTps":35563.31945301724,"stepMs":460.6480255126953,"jitter":0.00036484232069868824,"achievedTflops":300.08340101630665,"nominalPeakTflops":989.5,"mfuNominalPct":30.326771199222502,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.326771199222502,"vramAllocatedGb":64.927866368,"vramAllocatedPct":76.37000773840174,"vramReservedGb":66.225963008,"vramReservedPct":77.89686601956117,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.105955328,"vramAllocatedPct":98.9278228063062,"vramReservedGb":84.198555648,"vramReservedPct":99.03674194304321,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 30.19 MiB is free. Including non-PyTorch memory, this process has 79.14 GiB memory in use. Of the allocated memory 78.33 GiB is allocated by PyTorch, and 88.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":38185.42149700368,"meanTps":38181.94876354174,"stepMs":429.06427001953125,"jitter":0.001897753325170926,"achievedTflops":322.17277416505357,"nominalPeakTflops":989.5,"mfuNominalPct":32.559148475498084,"configuredPeakTflops":989.5,"mfuConfiguredPct":32.559148475498084,"vramAllocatedGb":64.927866368,"vramAllocatedPct":43.25042497285718,"vramReservedGb":66.211282944,"vramReservedPct":44.10534775769963,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":39068.514525356775,"meanTps":39051.85029748444,"stepMs":838.7316589355469,"jitter":0.0017064954506073975,"achievedTflops":329.6234849242018,"nominalPeakTflops":989.5,"mfuNominalPct":33.31212581346153,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.31212581346153,"vramAllocatedGb":115.842094592,"vramAllocatedPct":77.16593969764622,"vramReservedGb":118.3318016,"vramReservedPct":78.82440919891683,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.08873216,"vramAllocatedPct":99.31253536097222,"vramReservedGb":149.273182208,"vramReservedPct":99.43540314345947,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 59.44 MiB is free. Including non-PyTorch memory, this process has 139.74 GiB memory in use. Of the allocated memory 138.85 GiB is allocated by PyTorch, and 175.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":38242.89636575211,"meanTps":38181.92201610106,"stepMs":428.41943359375,"jitter":0.001446486883735296,"achievedTflops":322.65769320440234,"nominalPeakTflops":989.5,"mfuNominalPct":32.60815494738781,"configuredPeakTflops":989.5,"mfuConfiguredPct":32.60815494738781,"vramAllocatedGb":64.927866368,"vramAllocatedPct":43.25042497285718,"vramReservedGb":66.211282944,"vramReservedPct":44.10534775769963,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":39151.99377326325,"meanTps":39198.44310447316,"stepMs":836.9433288574219,"jitter":0.0019048127291639706,"achievedTflops":330.32780452652315,"nominalPeakTflops":989.5,"mfuNominalPct":33.383305156798706,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.383305156798706,"vramAllocatedGb":115.842094592,"vramAllocatedPct":77.16593969764622,"vramReservedGb":118.3318016,"vramReservedPct":78.82440919891683,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.08873216,"vramAllocatedPct":99.31253536097222,"vramReservedGb":149.273182208,"vramReservedPct":99.43540314345947,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 59.44 MiB is free. Including non-PyTorch memory, this process has 139.74 GiB memory in use. Of the allocated memory 138.85 GiB is allocated by PyTorch, and 175.91 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":10404.51214012081,"meanTps":10420.702940951576,"stepMs":393.67535400390625,"jitter":0.002777528834598945,"achievedTflops":87.7835155042017,"nominalPeakTflops":364.2,"mfuNominalPct":24.103106947886243,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":19.44550174056523,"vramAllocatedGb":26.692092928,"vramAllocatedPct":52.46413080512006,"vramReservedGb":26.83305984,"vramReservedPct":52.741205612641195,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10458.004619147106,"meanTps":10457.419250052128,"stepMs":783.3234252929688,"jitter":0.0023200189403575866,"achievedTflops":88.23483487398319,"nominalPeakTflops":364.2,"mfuNominalPct":24.227027697414382,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":19.54547645153721,"vramAllocatedGb":39.420682752,"vramAllocatedPct":77.48256616320096,"vramReservedGb":40.221278208,"vramReservedPct":79.05616119146897,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":50.104306688,"vramAllocatedPct":98.48155808050556,"vramReservedGb":50.287607808,"vramReservedPct":98.84184207780564,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 96.50 MiB is free. Process 2319972 has 47.28 GiB memory in use. Of the allocated memory 46.61 GiB is allocated by PyTorch, and 170.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7091.813967929157,"meanTps":7090.526407700008,"stepMs":577.5673217773438,"jitter":0.001007189272422099,"achievedTflops":59.83407516110552,"nominalPeakTflops":154.8,"mfuNominalPct":38.65250333404749,"configuredPeakTflops":180.6,"mfuConfiguredPct":33.13071714346928,"vramAllocatedGb":26.692092928,"vramAllocatedPct":52.302638654528735,"vramReservedGb":26.7911168,"vramReservedPct":52.4966740120917,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":7989.067732274678,"meanTps":7989.75883925638,"stepMs":1025.4012451171875,"jitter":0.0006107719241479285,"achievedTflops":67.40426092982662,"nominalPeakTflops":154.8,"mfuNominalPct":43.54280421823425,"configuredPeakTflops":180.6,"mfuConfiguredPct":37.32240361562936,"vramAllocatedGb":39.420682752,"vramAllocatedPct":77.24406366537994,"vramReservedGb":40.179335168,"vramReservedPct":78.73062836772328,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.542578688,"vramAllocatedPct":99.03720314915836,"vramReservedGb":50.6462208,"vramReservedPct":99.24028785847472,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 32.25 MiB is free. Process 1415479 has 47.49 GiB memory in use. Of the allocated memory 47.07 GiB is allocated by PyTorch, and 98.84 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7095.627651191184,"meanTps":7091.138640161808,"stepMs":577.2568969726562,"jitter":0.0022234821565363613,"achievedTflops":59.866251443784776,"nominalPeakTflops":154.8,"mfuNominalPct":38.673289046372595,"configuredPeakTflops":180.6,"mfuConfiguredPct":33.14853346831937,"vramAllocatedGb":26.692092928,"vramAllocatedPct":52.302638654528735,"vramReservedGb":26.7911168,"vramReservedPct":52.4966740120917,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":7998.207017386414,"meanTps":7997.450806644485,"stepMs":1024.2295532226562,"jitter":0.000556340205936758,"achievedTflops":67.48136964626606,"nominalPeakTflops":154.8,"mfuNominalPct":43.592616050559464,"configuredPeakTflops":180.6,"mfuConfiguredPct":37.36509947190812,"vramAllocatedGb":39.420682752,"vramAllocatedPct":77.24406366537994,"vramReservedGb":40.179335168,"vramReservedPct":78.73062836772328,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.542578688,"vramAllocatedPct":99.03720314915836,"vramReservedGb":50.6462208,"vramReservedPct":99.24028785847472,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 32.25 MiB is free. Process 1449495 has 47.49 GiB memory in use. Of the allocated memory 47.07 GiB is allocated by PyTorch, and 98.84 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":20708.225429558883,"meanTps":20700.14945014432,"stepMs":197.7957992553711,"jitter":0.0003384324254456743,"achievedTflops":174.7165848411494,"nominalPeakTflops":468.0,"mfuNominalPct":37.3326035985362,"configuredPeakTflops":468.0,"mfuConfiguredPct":37.3326035985362,"vramAllocatedGb":26.692092928,"vramAllocatedPct":26.175403430529197,"vramReservedGb":26.81208832,"vramReservedPct":26.29307602382778,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":23294.6186081629,"meanTps":23294.530168365578,"stepMs":351.66920471191406,"jitter":0.0004112580673756442,"achievedTflops":196.53814481784914,"nominalPeakTflops":468.0,"mfuNominalPct":41.995330089284,"configuredPeakTflops":468.0,"mfuConfiguredPct":41.995330089284,"vramAllocatedGb":39.420682752,"vramAllocatedPct":38.657600860444,"vramReservedGb":40.179335168,"vramReservedPct":39.40156773879675,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23168.44319002687,"meanTps":23167.752031880726,"stepMs":707.1687927246094,"jitter":0.0001145899578323571,"achievedTflops":195.47359497399032,"nominalPeakTflops":468.0,"mfuNominalPct":41.76786217392956,"configuredPeakTflops":468.0,"mfuConfiguredPct":41.76786217392956,"vramAllocatedGb":64.877796864,"vramAllocatedPct":63.62193145288011,"vramReservedGb":66.192408576,"vramReservedPct":64.91109570121831,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":20556.028070740766,"meanTps":20529.371625746233,"stepMs":199.26028442382812,"jitter":0.0011474812160026072,"achievedTflops":173.43248626664828,"nominalPeakTflops":468.0,"mfuNominalPct":37.058223561249626,"configuredPeakTflops":468.0,"mfuConfiguredPct":37.058223561249626,"vramAllocatedGb":26.740778496,"vramAllocatedPct":26.22314657255564,"vramReservedGb":28.294774784,"vramReservedPct":27.747061534101242,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":23182.232975453782,"meanTps":23193.74014860662,"stepMs":353.3740692138672,"jitter":0.002704357740277821,"achievedTflops":195.5899402505896,"nominalPeakTflops":468.0,"mfuNominalPct":41.79272227576701,"configuredPeakTflops":468.0,"mfuConfiguredPct":41.79272227576701,"vramAllocatedGb":39.467460096,"vramAllocatedPct":38.703472716724114,"vramReservedGb":41.672507392,"vramReservedPct":40.8658360320283,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23096.298240172357,"meanTps":23095.45122079882,"stepMs":709.3777465820312,"jitter":0.0009496466277104344,"achievedTflops":194.864903548692,"nominalPeakTflops":468.0,"mfuNominalPct":41.63779990356667,"configuredPeakTflops":468.0,"mfuConfiguredPct":41.63779990356667,"vramAllocatedGb":64.911189504,"vramAllocatedPct":63.654677698218315,"vramReservedGb":68.769808384,"vramReservedPct":67.43860375231604,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":12981.937613586018,"meanTps":12983.92234875489,"stepMs":631.0306091308594,"jitter":0.0032426659443395077,"achievedTflops":141.5461277972799,"nominalPeakTflops":312.0,"mfuNominalPct":45.36734865297433,"configuredPeakTflops":312.0,"mfuConfiguredPct":45.36734865297433,"vramAllocatedGb":39.43561728,"vramAllocatedPct":46.40890862784628,"vramReservedGb":39.577452544,"vramReservedPct":46.57582423006562,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":13676.243935662076,"meanTps":13680.644731529137,"stepMs":1197.98974609375,"jitter":0.0014643306882501325,"achievedTflops":149.11636687254608,"nominalPeakTflops":312.0,"mfuNominalPct":47.79370733094426,"configuredPeakTflops":312.0,"mfuConfiguredPct":47.79370733094426,"vramAllocatedGb":64.893116928,"vramAllocatedPct":76.36798766720602,"vramReservedGb":66.2175744,"vramReservedPct":77.92664529802468,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.344178688,"vramAllocatedPct":99.25852698665128,"vramReservedGb":84.422950912,"vramReservedPct":99.35122828558295,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 768.00 KiB is free. Process 764857 has 79.13 GiB memory in use. Of the allocated memory 78.55 GiB is allocated by PyTorch, and 75.12 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":12975.360077896014,"meanTps":12967.869487034399,"stepMs":631.3504943847656,"jitter":0.001229689628715416,"achievedTflops":141.4744108675672,"nominalPeakTflops":312.0,"mfuNominalPct":45.34436245755359,"configuredPeakTflops":312.0,"mfuConfiguredPct":45.34436245755359,"vramAllocatedGb":39.43561728,"vramAllocatedPct":46.40890862784628,"vramReservedGb":39.577452544,"vramReservedPct":46.57582423006562,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":13677.659933092264,"meanTps":13675.042658307015,"stepMs":1197.86572265625,"jitter":0.0014181742613510603,"achievedTflops":149.13180593558738,"nominalPeakTflops":312.0,"mfuNominalPct":47.7986557485857,"configuredPeakTflops":312.0,"mfuConfiguredPct":47.7986557485857,"vramAllocatedGb":64.893116928,"vramAllocatedPct":76.36798766720602,"vramReservedGb":66.2175744,"vramReservedPct":77.92664529802468,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.344178688,"vramAllocatedPct":99.25852698665128,"vramReservedGb":84.422950912,"vramReservedPct":99.35122828558295,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 768.00 KiB is free. Process 852916 has 79.13 GiB memory in use. Of the allocated memory 78.55 GiB is allocated by PyTorch, and 75.12 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":40213.69256415729,"meanTps":40114.86489147115,"stepMs":407.4234161376953,"jitter":0.0012968703041144566,"achievedTflops":438.4624727305555,"nominalPeakTflops":2250.0,"mfuNominalPct":19.48722101024691,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.48722101024691,"vramAllocatedGb":64.893116928,"vramAllocatedPct":33.88621290729214,"vramReservedGb":66.184019968,"vramReservedPct":34.56030312405034,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":42978.30166630595,"meanTps":42960.63874896195,"stepMs":762.4312438964844,"jitter":0.0002729802375958674,"achievedTflops":468.605870806412,"nominalPeakTflops":2250.0,"mfuNominalPct":20.82692759139609,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.82692759139609,"vramAllocatedGb":115.807985152,"vramAllocatedPct":60.4731938763131,"vramReservedGb":118.304538624,"vramReservedPct":61.77685667587464,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.086769152,"vramAllocatedPct":98.7382764268486,"vramReservedGb":189.794353152,"vramReservedPct":99.10776618491333,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.19 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.19 GiB is free. Including non-PyTorch memory, this process has 177.15 GiB memory in use. Of the allocated memory 176.10 GiB is allocated by PyTorch, and 234.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":40283.19910500273,"meanTps":40110.16900732842,"stepMs":406.7204284667969,"jitter":0.0015660427397210935,"achievedTflops":439.22032429371194,"nominalPeakTflops":2250.0,"mfuNominalPct":19.520903301942752,"configuredPeakTflops":2250.0,"mfuConfiguredPct":19.520903301942752,"vramAllocatedGb":64.893116928,"vramAllocatedPct":33.88621290729214,"vramReservedGb":66.184019968,"vramReservedPct":34.56030312405034,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":42956.03388949836,"meanTps":42801.23365913688,"stepMs":762.8264770507812,"jitter":0.0011167410790189811,"achievedTflops":468.3630782683805,"nominalPeakTflops":2250.0,"mfuNominalPct":20.816136811928022,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.816136811928022,"vramAllocatedGb":115.807985152,"vramAllocatedPct":60.4731938763131,"vramReservedGb":118.304538624,"vramReservedPct":61.77685667587464,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.086769152,"vramAllocatedPct":98.7382764268486,"vramReservedGb":189.794353152,"vramReservedPct":99.10776618491333,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.19 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.19 GiB is free. Including non-PyTorch memory, this process has 177.15 GiB memory in use. Of the allocated memory 176.10 GiB is allocated by PyTorch, and 234.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":42260.11466894679,"meanTps":42260.53680984348,"stepMs":775.3883361816406,"jitter":0.0002881415463875936,"achievedTflops":460.7752532563664,"nominalPeakTflops":2250.0,"mfuNominalPct":20.478900144727398,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.478900144727398,"vramAllocatedGb":115.807985152,"vramAllocatedPct":40.29103879229461,"vramReservedGb":118.26888704,"vramReservedPct":41.14721717501407,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":44776.46310161989,"meanTps":44769.00864782894,"stepMs":1463.6260986328125,"jitter":0.00020854042422763922,"achievedTflops":488.21178757316034,"nominalPeakTflops":2250.0,"mfuNominalPct":21.698301669918237,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.698301669918237,"vramAllocatedGb":217.6377216,"vramAllocatedPct":75.71887095818951,"vramReservedGb":223.006949376,"vramReservedPct":77.58689209959476,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.01660416,"vramAllocatedPct":99.50873489704189,"vramReservedGb":286.30319104,"vramReservedPct":99.60844203799843,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 739.62 MiB is free. Including non-PyTorch memory, this process has 266.95 GiB memory in use. Of the allocated memory 265.94 GiB is allocated by PyTorch, and 201.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":42264.602634725525,"meanTps":42244.76664501725,"stepMs":775.3059997558594,"jitter":0.000588949636230371,"achievedTflops":460.82418695151836,"nominalPeakTflops":2250.0,"mfuNominalPct":20.481074975623038,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.481074975623038,"vramAllocatedGb":115.807985152,"vramAllocatedPct":40.29103879229461,"vramReservedGb":118.26888704,"vramReservedPct":41.14721717501407,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":8,"tokensPerStep":65536,"status":"complete","stable":true,"tps":44775.32598249281,"meanTps":44769.69071986562,"stepMs":1463.6632690429688,"jitter":0.0003599441199676789,"achievedTflops":488.1993892075177,"nominalPeakTflops":2250.0,"mfuNominalPct":21.69775063144523,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.69775063144523,"vramAllocatedGb":217.6377216,"vramAllocatedPct":75.71887095818951,"vramReservedGb":223.006949376,"vramReservedPct":77.58689209959476,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.01660416,"vramAllocatedPct":99.50873489704189,"vramReservedGb":286.30319104,"vramReservedPct":99.60844203799843,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 739.62 MiB is free. Including non-PyTorch memory, this process has 266.95 GiB memory in use. Of the allocated memory 265.94 GiB is allocated by PyTorch, and 201.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.717022208,"vramAllocatedPct":97.84638545425771,"vramReservedGb":24.744296448,"vramReservedPct":97.95435500566866,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 53.69 MiB is free. Including non-PyTorch memory, this process has 23.46 GiB memory in use. Of the allocated memory 22.99 GiB is allocated by PyTorch, and 14.01 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.717022208,"vramAllocatedPct":97.84638545425771,"vramReservedGb":24.744296448,"vramReservedPct":97.95435500566866,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 53.69 MiB is free. Including non-PyTorch memory, this process has 23.46 GiB memory in use. Of the allocated memory 22.99 GiB is allocated by PyTorch, and 14.01 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed11.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.964448256,"vramAllocatedPct":97.90783042593033,"vramReservedGb":33.01965824,"vramReservedPct":98.0718098048452,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 9.88 MiB is free. Including non-PyTorch memory, this process has 31.34 GiB memory in use. Of the allocated memory 30.69 GiB is allocated by PyTorch, and 59.71 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_context_frontier_seed22.json","sourceKind":"context_frontier","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.964448256,"vramAllocatedPct":97.90783042593033,"vramReservedGb":33.01965824,"vramReservedPct":98.0718098048452,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 9.88 MiB is free. Including non-PyTorch memory, this process has 31.34 GiB memory in use. Of the allocated memory 30.69 GiB is allocated by PyTorch, and 59.71 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":27804.75313541602,"meanTps":27804.958771645543,"stepMs":294.6258850097656,"jitter":0.00034998343225470176,"achievedTflops":303.16392343147794,"nominalPeakTflops":989.5,"mfuNominalPct":30.638092312428288,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.638092312428288,"vramAllocatedGb":39.485686784,"vramAllocatedPct":46.44419066782089,"vramReservedGb":39.619395584,"vramReservedPct":46.60146277087779,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":30198.73192203659,"meanTps":30198.443321743478,"stepMs":542.5393371582031,"jitter":0.0004898654055823938,"achievedTflops":329.2662232083884,"nominalPeakTflops":989.5,"mfuNominalPct":33.27602053647179,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.27602053647179,"vramAllocatedGb":64.943186432,"vramAllocatedPct":76.38802763450616,"vramReservedGb":66.23854592,"vramReservedPct":77.91166639943759,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.120635392,"vramAllocatedPct":98.94508991616202,"vramReservedGb":84.213235712,"vramReservedPct":99.05400905289902,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 16.19 MiB is free. Including non-PyTorch memory, this process has 79.15 GiB memory in use. Of the allocated memory 78.34 GiB is allocated by PyTorch, and 88.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":27780.406471772767,"meanTps":27776.69247295183,"stepMs":294.88409423828125,"jitter":0.0007300408309756568,"achievedTflops":302.8984641398019,"nominalPeakTflops":989.5,"mfuNominalPct":30.61126469325941,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.61126469325941,"vramAllocatedGb":39.485686784,"vramAllocatedPct":46.44419066782089,"vramReservedGb":39.619395584,"vramReservedPct":46.60146277087779,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":30190.113661336407,"meanTps":30190.522471783243,"stepMs":542.6942138671875,"jitter":0.0005008140469405507,"achievedTflops":329.172255615355,"nominalPeakTflops":989.5,"mfuNominalPct":33.266524064209705,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.266524064209705,"vramAllocatedGb":64.943186432,"vramAllocatedPct":76.38802763450616,"vramReservedGb":66.23854592,"vramReservedPct":77.91166639943759,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.120635392,"vramAllocatedPct":98.94508991616202,"vramReservedGb":84.213235712,"vramReservedPct":99.05400905289902,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 16.19 MiB is free. Including non-PyTorch memory, this process has 79.15 GiB memory in use. Of the allocated memory 78.34 GiB is allocated by PyTorch, and 88.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":32018.258970540803,"meanTps":32038.36323562559,"stepMs":511.70802307128906,"jitter":0.0020903878024399187,"achievedTflops":349.1050959409652,"nominalPeakTflops":989.5,"mfuNominalPct":35.28095967063822,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.28095967063822,"vramAllocatedGb":64.943186432,"vramAllocatedPct":43.26063013307076,"vramReservedGb":66.225963008,"vramReservedPct":44.11512659446335,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":33135.64201099991,"meanTps":33127.76123920731,"stepMs":988.9049377441406,"jitter":0.0012655278041090485,"achievedTflops":361.28827285577466,"nominalPeakTflops":989.5,"mfuNominalPct":36.512205442726085,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.512205442726085,"vramAllocatedGb":115.858054656,"vramAllocatedPct":77.17657118130967,"vramReservedGb":118.346481664,"vramReservedPct":78.83418803568053,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.103444992,"vramAllocatedPct":99.32233602549657,"vramReservedGb":149.28576512,"vramReservedPct":99.44378500354264,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 47.44 MiB is free. Including non-PyTorch memory, this process has 139.76 GiB memory in use. Of the allocated memory 138.86 GiB is allocated by PyTorch, and 173.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":32050.76916312543,"meanTps":32029.676467705962,"stepMs":511.18898010253906,"jitter":0.002785898025809819,"achievedTflops":349.4595647430246,"nominalPeakTflops":989.5,"mfuNominalPct":35.31678269257449,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.31678269257449,"vramAllocatedGb":64.943186432,"vramAllocatedPct":43.26063013307076,"vramReservedGb":66.225963008,"vramReservedPct":44.11512659446335,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":33162.242879861675,"meanTps":33140.673794612434,"stepMs":988.1116943359375,"jitter":0.0009072775825793234,"achievedTflops":361.57831045228,"nominalPeakTflops":989.5,"mfuNominalPct":36.54151697344921,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.54151697344921,"vramAllocatedGb":115.858054656,"vramAllocatedPct":77.17657118130967,"vramReservedGb":118.346481664,"vramReservedPct":78.83418803568053,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.103444992,"vramAllocatedPct":99.32233602549657,"vramReservedGb":149.28576512,"vramReservedPct":99.44378500354264,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 47.44 MiB is free. Including non-PyTorch memory, this process has 139.76 GiB memory in use. Of the allocated memory 138.86 GiB is allocated by PyTorch, and 173.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":9636.55652479585,"meanTps":9639.501775547355,"stepMs":850.09619140625,"jitter":0.0023522796707863367,"achievedTflops":105.07039102983953,"nominalPeakTflops":364.2,"mfuNominalPct":28.849640590290917,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":23.27483081438929,"vramAllocatedGb":39.43561728,"vramAllocatedPct":77.51192043799006,"vramReservedGb":39.577452544,"vramReservedPct":77.79070201811369,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":50.177772544,"vramAllocatedPct":98.62595748332838,"vramReservedGb":50.291802112,"vramReservedPct":98.85008611150829,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 32.50 MiB is free. Process 2319972 has 47.34 GiB memory in use. Of the allocated memory 46.73 GiB is allocated by PyTorch, and 108.75 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":7201.265581343567,"meanTps":7200.644693398499,"stepMs":1137.5778198242188,"jitter":0.0010935304316567027,"achievedTflops":78.51765188058425,"nominalPeakTflops":154.8,"mfuNominalPct":50.721997338878715,"configuredPeakTflops":180.6,"mfuConfiguredPct":43.4759977190389,"vramAllocatedGb":39.43561728,"vramAllocatedPct":77.27332758348358,"vramReservedGb":39.577452544,"vramReservedPct":77.551251033753,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.488019968,"vramAllocatedPct":98.930296394963,"vramReservedGb":50.629443584,"vramReservedPct":99.20741322895638,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 48.25 MiB is free. Process 1415479 has 47.47 GiB memory in use. Of the allocated memory 47.02 GiB is allocated by PyTorch, and 134.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":7207.293805993464,"meanTps":7206.34329159774,"stepMs":1136.6263427734375,"jitter":0.0006682751379166025,"achievedTflops":78.58337950014945,"nominalPeakTflops":154.8,"mfuNominalPct":50.76445704144021,"configuredPeakTflops":180.6,"mfuConfiguredPct":43.5123917498059,"vramAllocatedGb":39.43561728,"vramAllocatedPct":77.27332758348358,"vramReservedGb":39.577452544,"vramReservedPct":77.551251033753,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.488019968,"vramAllocatedPct":98.930296394963,"vramReservedGb":50.629443584,"vramReservedPct":99.20741322895638,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 48.25 MiB is free. Process 1449495 has 47.47 GiB memory in use. Of the allocated memory 47.02 GiB is allocated by PyTorch, and 134.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":21188.04694190072,"meanTps":21180.692159320497,"stepMs":386.6330871582031,"jitter":0.0007054793545255802,"achievedTflops":231.0199054626795,"nominalPeakTflops":468.0,"mfuNominalPct":49.363227662965706,"configuredPeakTflops":468.0,"mfuConfiguredPct":49.363227662965706,"vramAllocatedGb":39.43561728,"vramAllocatedPct":38.6722462948241,"vramReservedGb":39.556481024,"vramReservedPct":38.79077043108639,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21123.388914974876,"meanTps":21122.55675959953,"stepMs":775.6331176757812,"jitter":0.0002870589432319379,"achievedTflops":230.314918763869,"nominalPeakTflops":468.0,"mfuNominalPct":49.21258947945918,"configuredPeakTflops":468.0,"mfuConfiguredPct":49.21258947945918,"vramAllocatedGb":64.893116928,"vramAllocatedPct":63.63695496028596,"vramReservedGb":66.19660288,"vramReservedPct":64.91520881440155,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":21106.791447626136,"meanTps":21102.77165589108,"stepMs":388.12152099609375,"jitter":0.0005022644422858183,"achievedTflops":230.13395138408455,"nominalPeakTflops":468.0,"mfuNominalPct":49.1739212359155,"configuredPeakTflops":468.0,"mfuConfiguredPct":49.1739212359155,"vramAllocatedGb":39.4995072,"vramAllocatedPct":38.734899472139766,"vramReservedGb":42.186309632,"vramReservedPct":41.36969239697455,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21085.953303845963,"meanTps":21084.42876324819,"stepMs":777.0101623535156,"jitter":0.0007549529440417447,"achievedTflops":229.9067465822775,"nominalPeakTflops":468.0,"mfuNominalPct":49.12537320134134,"configuredPeakTflops":468.0,"mfuConfiguredPct":49.12537320134134,"vramAllocatedGb":64.934782464,"vramAllocatedPct":63.67781395987401,"vramReservedGb":67.310190592,"vramReservedPct":66.00724036455037,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":10918.34881867803,"meanTps":10920.345648866281,"stepMs":1500.5932006835938,"jitter":0.0005319354133999498,"achievedTflops":172.90094544704453,"nominalPeakTflops":312.0,"mfuNominalPct":55.41696969456556,"configuredPeakTflops":312.0,"mfuConfiguredPct":55.41696969456556,"vramAllocatedGb":64.922665984,"vramAllocatedPct":76.40276180121302,"vramReservedGb":65.183678464,"vramReservedPct":76.7099283975678,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.235061248,"vramAllocatedPct":99.1301146109375,"vramReservedGb":84.391493632,"vramReservedPct":99.3142085015934,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 90.75 MiB is free. Process 925444 has 79.04 GiB memory in use. Of the allocated memory 78.45 GiB is allocated by PyTorch, and 89.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":10919.491145702379,"meanTps":10907.577363100872,"stepMs":1500.4362182617188,"jitter":0.0010778533440002616,"achievedTflops":172.91903512578622,"nominalPeakTflops":312.0,"mfuNominalPct":55.422767668521224,"configuredPeakTflops":312.0,"mfuConfiguredPct":55.422767668521224,"vramAllocatedGb":64.922665984,"vramAllocatedPct":76.40276180121302,"vramReservedGb":65.183678464,"vramReservedPct":76.7099283975678,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.235061248,"vramAllocatedPct":99.1301146109375,"vramReservedGb":84.391493632,"vramReservedPct":99.3142085015934,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 90.75 MiB is free. Process 937161 has 79.04 GiB memory in use. Of the allocated memory 78.45 GiB is allocated by PyTorch, and 89.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":29327.607120532288,"meanTps":29273.566612406485,"stepMs":558.6545104980469,"jitter":0.0007330393967745313,"achievedTflops":464.4265431568677,"nominalPeakTflops":2250.0,"mfuNominalPct":20.641179695860785,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.641179695860785,"vramAllocatedGb":64.922665984,"vramAllocatedPct":33.90164298139286,"vramReservedGb":65.162706944,"vramReservedPct":34.02698877247353,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30710.027066786974,"meanTps":30710.31744731072,"stepMs":1067.0130615234375,"jitter":0.0007362988162299131,"achievedTflops":486.31828884861494,"nominalPeakTflops":2250.0,"mfuNominalPct":21.614146171049555,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.614146171049555,"vramAllocatedGb":115.83798528,"vramAllocatedPct":60.48885949345061,"vramReservedGb":118.3318016,"vramReservedPct":61.79109299326787,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.11612928,"vramAllocatedPct":98.75360784557977,"vramReservedGb":189.825810432,"vramReservedPct":99.12419270498243,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.19 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.16 GiB is free. Including non-PyTorch memory, this process has 177.18 GiB memory in use. Of the allocated memory 176.13 GiB is allocated by PyTorch, and 236.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":29456.749590620075,"meanTps":29454.655445032768,"stepMs":556.2052917480469,"jitter":0.00032920934052865357,"achievedTflops":466.4716193443355,"nominalPeakTflops":2250.0,"mfuNominalPct":20.732071970859355,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.732071970859355,"vramAllocatedGb":64.922665984,"vramAllocatedPct":33.90164298139286,"vramReservedGb":65.162706944,"vramReservedPct":34.02698877247353,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30762.035173872504,"meanTps":30758.73853463767,"stepMs":1065.2091064453125,"jitter":0.00022563021468812342,"achievedTflops":487.1418795797167,"nominalPeakTflops":2250.0,"mfuNominalPct":21.650750203542966,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.650750203542966,"vramAllocatedGb":115.83798528,"vramAllocatedPct":60.48885949345061,"vramReservedGb":118.3318016,"vramReservedPct":61.79109299326787,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.11612928,"vramAllocatedPct":98.75360784557977,"vramReservedGb":189.825810432,"vramReservedPct":99.12419270498243,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.19 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.16 GiB is free. Including non-PyTorch memory, this process has 177.18 GiB memory in use. Of the allocated memory 176.13 GiB is allocated by PyTorch, and 236.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30482.66005342391,"meanTps":30472.635523860507,"stepMs":1074.9718017578125,"jitter":0.0002198771792819329,"achievedTflops":482.7177470243152,"nominalPeakTflops":2250.0,"mfuNominalPct":21.454122089969562,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.454122089969562,"vramAllocatedGb":115.83798528,"vramAllocatedPct":40.301476210055014,"vramReservedGb":118.296150016,"vramReservedPct":41.15670230354099,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":31838.39617353975,"meanTps":31834.428757326532,"stepMs":2058.3951416015625,"jitter":9.98811733081696e-05,"achievedTflops":504.18693259784567,"nominalPeakTflops":2250.0,"mfuNominalPct":22.408308115459807,"configuredPeakTflops":2250.0,"mfuConfiguredPct":22.408308115459807,"vramAllocatedGb":217.668361728,"vramAllocatedPct":75.7295310399121,"vramReservedGb":223.038406656,"vramReservedPct":77.59783647866428,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.045964288,"vramAllocatedPct":99.51894965084011,"vramReservedGb":286.326259712,"vramReservedPct":99.61646791598274,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 717.62 MiB is free. Including non-PyTorch memory, this process has 266.97 GiB memory in use. Of the allocated memory 265.96 GiB is allocated by PyTorch, and 195.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30459.540854703922,"meanTps":30427.722758966775,"stepMs":1075.7877197265625,"jitter":0.0002625253479208152,"achievedTflops":482.3516356843087,"nominalPeakTflops":2250.0,"mfuNominalPct":21.437850474858166,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.437850474858166,"vramAllocatedGb":115.83798528,"vramAllocatedPct":40.301476210055014,"vramReservedGb":118.296150016,"vramReservedPct":41.15670230354099,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":4,"tokensPerStep":65536,"status":"complete","stable":true,"tps":31834.12956570119,"meanTps":31828.709036524906,"stepMs":2058.6710205078125,"jitter":0.00010980182017598603,"achievedTflops":504.11936738799665,"nominalPeakTflops":2250.0,"mfuNominalPct":22.405305217244297,"configuredPeakTflops":2250.0,"mfuConfiguredPct":22.405305217244297,"vramAllocatedGb":217.668361728,"vramAllocatedPct":75.7295310399121,"vramReservedGb":223.038406656,"vramReservedPct":77.59783647866428,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.045964288,"vramAllocatedPct":99.51894965084011,"vramReservedGb":286.326259712,"vramReservedPct":99.61646791598274,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 717.62 MiB is free. Including non-PyTorch memory, this process has 266.97 GiB memory in use. Of the allocated memory 265.96 GiB is allocated by PyTorch, and 195.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23454.582430111146,"meanTps":23447.743499076445,"stepMs":698.5415344238281,"jitter":0.0007038036708825312,"achievedTflops":371.4224141927411,"nominalPeakTflops":989.5,"mfuNominalPct":37.536373339337146,"configuredPeakTflops":989.5,"mfuConfiguredPct":37.536373339337146,"vramAllocatedGb":64.972735488,"vramAllocatedPct":76.42278407671839,"vramReservedGb":65.225621504,"vramReservedPct":76.72023581938603,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.031637504,"vramAllocatedPct":98.84040806266111,"vramReservedGb":84.2006528,"vramReservedPct":99.03920867302261,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 88.19 MiB is free. Including non-PyTorch memory, this process has 79.08 GiB memory in use. Of the allocated memory 78.26 GiB is allocated by PyTorch, and 101.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23477.35124046632,"meanTps":23479.12709278406,"stepMs":697.8640747070312,"jitter":0.0002706173666596553,"achievedTflops":371.78297684763436,"nominalPeakTflops":989.5,"mfuNominalPct":37.572812212999935,"configuredPeakTflops":989.5,"mfuConfiguredPct":37.572812212999935,"vramAllocatedGb":64.972735488,"vramAllocatedPct":76.42278407671839,"vramReservedGb":65.225621504,"vramReservedPct":76.72023581938603,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.031637504,"vramAllocatedPct":98.84040806266111,"vramReservedGb":84.2006528,"vramReservedPct":99.03920867302261,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 88.19 MiB is free. Including non-PyTorch memory, this process has 79.08 GiB memory in use. Of the allocated memory 78.26 GiB is allocated by PyTorch, and 101.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":24593.621500928024,"meanTps":24605.809388408146,"stepMs":666.1889953613281,"jitter":0.002120087856528916,"achievedTflops":389.4600255125456,"nominalPeakTflops":989.5,"mfuNominalPct":39.35927493810466,"configuredPeakTflops":989.5,"mfuConfiguredPct":39.35927493810466,"vramAllocatedGb":64.972735488,"vramAllocatedPct":43.28031365728059,"vramReservedGb":65.204649984,"vramReservedPct":43.43479895104514,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":25145.137409767205,"meanTps":25147.050626523833,"stepMs":1303.154541015625,"jitter":0.0008731353886799254,"achievedTflops":398.1937290835702,"nominalPeakTflops":989.5,"mfuNominalPct":40.24191299480245,"configuredPeakTflops":989.5,"mfuConfiguredPct":40.24191299480245,"vramAllocatedGb":115.888054784,"vramAllocatedPct":77.19655517828697,"vramReservedGb":118.39471616,"vramReservedPct":78.86631849933272,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.132870656,"vramAllocatedPct":99.34193735454527,"vramReservedGb":149.3172224,"vramReservedPct":99.4647396537506,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 17.44 MiB is free. Including non-PyTorch memory, this process has 139.79 GiB memory in use. Of the allocated memory 138.89 GiB is allocated by PyTorch, and 175.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":24591.332434610147,"meanTps":24587.186502054603,"stepMs":666.2510070800781,"jitter":0.003791436704943858,"achievedTflops":389.42377628318644,"nominalPeakTflops":989.5,"mfuNominalPct":39.355611549589334,"configuredPeakTflops":989.5,"mfuConfiguredPct":39.355611549589334,"vramAllocatedGb":64.972735488,"vramAllocatedPct":43.28031365728059,"vramReservedGb":65.204649984,"vramReservedPct":43.43479895104514,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":25152.924479886267,"meanTps":25147.786475096826,"stepMs":1302.7510986328125,"jitter":0.0007377479599446671,"achievedTflops":398.31704368069546,"nominalPeakTflops":989.5,"mfuNominalPct":40.25437530881208,"configuredPeakTflops":989.5,"mfuConfiguredPct":40.25437530881208,"vramAllocatedGb":115.888054784,"vramAllocatedPct":77.19655517828697,"vramReservedGb":118.39471616,"vramReservedPct":78.86631849933272,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.132870656,"vramAllocatedPct":99.34193735454527,"vramReservedGb":149.3172224,"vramReservedPct":99.4647396537506,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 17.44 MiB is free. Including non-PyTorch memory, this process has 139.79 GiB memory in use. Of the allocated memory 138.89 GiB is allocated by PyTorch, and 175.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.496922624,"vramAllocatedPct":98.94774097681311,"vramReservedGb":50.564431872,"vramReservedPct":99.08002403957283,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 110.25 MiB is free. Process 1472317 has 47.41 GiB memory in use. Of the allocated memory 47.03 GiB is allocated by PyTorch, and 64.38 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.496922624,"vramAllocatedPct":98.94774097681311,"vramReservedGb":50.564431872,"vramReservedPct":99.08002403957283,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 110.25 MiB is free. Process 1479887 has 47.41 GiB memory in use. Of the allocated memory 47.03 GiB is allocated by PyTorch, and 64.38 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":18003.57871113884,"meanTps":18004.101683339566,"stepMs":910.0412902832031,"jitter":0.000174957365356602,"achievedTflops":285.10133100538593,"nominalPeakTflops":468.0,"mfuNominalPct":60.919087821663666,"configuredPeakTflops":468.0,"mfuConfiguredPct":60.919087821663666,"vramAllocatedGb":64.979609088,"vramAllocatedPct":63.72177285701983,"vramReservedGb":68.633493504,"vramReservedPct":67.30492757386091,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.922630144,"vramAllocatedPct":98.96903050708902,"vramReservedGb":101.19806976,"vramReservedPct":99.23913832849507,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 71.75 MiB is free. Including non-PyTorch memory, this process has 94.89 GiB memory in use. Of the allocated memory 93.99 GiB is allocated by PyTorch, and 262.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":18011.98544670819,"meanTps":18012.063583164218,"stepMs":909.6165466308594,"jitter":0.00021403611281703221,"achievedTflops":285.234458509572,"nominalPeakTflops":468.0,"mfuNominalPct":60.94753386956667,"configuredPeakTflops":468.0,"mfuConfiguredPct":60.94753386956667,"vramAllocatedGb":64.979609088,"vramAllocatedPct":63.72177285701983,"vramReservedGb":68.633493504,"vramReservedPct":67.30492757386091,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.922630144,"vramAllocatedPct":98.96903050708902,"vramReservedGb":101.19806976,"vramReservedPct":99.23913832849507,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 71.75 MiB is free. Including non-PyTorch memory, this process has 94.89 GiB memory in use. Of the allocated memory 93.99 GiB is allocated by PyTorch, and 262.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.8804992,"vramAllocatedPct":98.71285633470204,"vramReservedGb":83.961577472,"vramReservedPct":98.80827145373607,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 440.75 MiB is free. Process 949988 has 78.70 GiB memory in use. Of the allocated memory 78.12 GiB is allocated by PyTorch, and 77.32 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":83.8804992,"vramAllocatedPct":98.71285633470204,"vramReservedGb":83.961577472,"vramReservedPct":98.80827145373607,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 440.75 MiB is free. Process 970883 has 78.70 GiB memory in use. Of the allocated memory 78.12 GiB is allocated by PyTorch, and 77.32 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":19709.526704166114,"meanTps":19677.087545880902,"stepMs":1662.5462646484375,"jitter":0.0013091261925203413,"achievedTflops":506.5509470218733,"nominalPeakTflops":2250.0,"mfuNominalPct":22.51337542319437,"configuredPeakTflops":2250.0,"mfuConfiguredPct":22.51337542319437,"vramAllocatedGb":115.896763392,"vramAllocatedPct":60.519552542448864,"vramReservedGb":116.2870784,"vramReservedPct":60.7233691887763,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.174849536,"vramAllocatedPct":98.78427068304208,"vramReservedGb":189.966319616,"vramReservedPct":99.19756449462442,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.19 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.12 GiB is free. Including non-PyTorch memory, this process has 177.21 GiB memory in use. Of the allocated memory 176.18 GiB is allocated by PyTorch, and 214.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":19637.534603474054,"meanTps":19591.343235561166,"stepMs":1668.6412353515625,"jitter":0.0054835936023744525,"achievedTflops":504.70069118716833,"nominalPeakTflops":2250.0,"mfuNominalPct":22.431141830540817,"configuredPeakTflops":2250.0,"mfuConfiguredPct":22.431141830540817,"vramAllocatedGb":115.896763392,"vramAllocatedPct":60.519552542448864,"vramReservedGb":116.2870784,"vramReservedPct":60.7233691887763,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":189.174849536,"vramAllocatedPct":98.78427068304208,"vramReservedGb":189.966319616,"vramReservedPct":99.19756449462442,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.19 GiB. GPU 0 has a total capacity of 178.35 GiB of which 1.12 GiB is free. Including non-PyTorch memory, this process has 177.21 GiB memory in use. Of the allocated memory 176.18 GiB is allocated by PyTorch, and 214.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":19711.51094182142,"meanTps":19710.296978017202,"stepMs":1662.37890625,"jitter":0.00020587118363884585,"achievedTflops":506.6019435515464,"nominalPeakTflops":2250.0,"mfuNominalPct":22.515641935624284,"configuredPeakTflops":2250.0,"mfuConfiguredPct":22.515641935624284,"vramAllocatedGb":115.896763392,"vramAllocatedPct":40.32192584647362,"vramReservedGb":116.2870784,"vramReservedPct":40.45772129363473,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":20278.122282380427,"meanTps":20278.09233480158,"stepMs":3231.857421875,"jitter":7.470293906307505e-05,"achievedTflops":521.1643181565555,"nominalPeakTflops":2250.0,"mfuNominalPct":23.1628585847358,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.1628585847358,"vramAllocatedGb":217.727721984,"vramAllocatedPct":75.75018321147071,"vramReservedGb":223.187304448,"vramReservedPct":77.64963987292664,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.104684544,"vramAllocatedPct":99.53937915843653,"vramReservedGb":286.454185984,"vramReservedPct":99.6609750575321,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 535.62 MiB is free. Including non-PyTorch memory, this process has 267.15 GiB memory in use. Of the allocated memory 266.02 GiB is allocated by PyTorch, and 321.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":19723.796486563962,"meanTps":19723.840616105503,"stepMs":1661.3434448242188,"jitter":0.00012239049935283412,"achievedTflops":506.9176920937324,"nominalPeakTflops":2250.0,"mfuNominalPct":22.529675204165883,"configuredPeakTflops":2250.0,"mfuConfiguredPct":22.529675204165883,"vramAllocatedGb":115.896763392,"vramAllocatedPct":40.32192584647362,"vramReservedGb":116.2870784,"vramReservedPct":40.45772129363473,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":2,"tokensPerStep":65536,"status":"complete","stable":true,"tps":20284.09824911809,"meanTps":20283.33527450344,"stepMs":3230.9052734375,"jitter":0.00013690238955014239,"achievedTflops":521.3179053865165,"nominalPeakTflops":2250.0,"mfuNominalPct":23.169684683845176,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.169684683845176,"vramAllocatedGb":217.727721984,"vramAllocatedPct":75.75018321147071,"vramReservedGb":223.187304448,"vramReservedPct":77.64963987292664,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.104684544,"vramAllocatedPct":99.53937915843653,"vramReservedGb":286.454185984,"vramReservedPct":99.6609750575321,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 535.62 MiB is free. Including non-PyTorch memory, this process has 267.15 GiB memory in use. Of the allocated memory 266.02 GiB is allocated by PyTorch, and 321.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_32k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.930568704,"vramAllocatedPct":98.72152805828266,"vramReservedGb":84.066435072,"vramReservedPct":98.88133795434082,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 196.19 MiB is free. Including non-PyTorch memory, this process has 78.98 GiB memory in use. Of the allocated memory 78.17 GiB is allocated by PyTorch, and 89.57 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_32k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.930568704,"vramAllocatedPct":98.72152805828266,"vramReservedGb":84.066435072,"vramReservedPct":98.88133795434082,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 608.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 196.19 MiB is free. Including non-PyTorch memory, this process has 78.98 GiB memory in use. Of the allocated memory 78.17 GiB is allocated by PyTorch, and 89.57 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":16903.73943317359,"meanTps":16904.274283533465,"stepMs":1938.5059814453125,"jitter":0.0006918539397481435,"achievedTflops":434.4399206844039,"nominalPeakTflops":989.5,"mfuNominalPct":43.90499451080383,"configuredPeakTflops":989.5,"mfuConfiguredPct":43.90499451080383,"vramAllocatedGb":115.946832896,"vramAllocatedPct":77.23570906498168,"vramReservedGb":116.391936,"vramReservedPct":77.53220576942638,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.074281472,"vramAllocatedPct":99.30290931853295,"vramReservedGb":149.268987904,"vramReservedPct":99.4326091900984,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 65.44 MiB is free. Including non-PyTorch memory, this process has 139.74 GiB memory in use. Of the allocated memory 138.84 GiB is allocated by PyTorch, and 183.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":16906.656000001007,"meanTps":16907.922004168897,"stepMs":1938.1715698242188,"jitter":0.0006842229868471435,"achievedTflops":434.51487883589357,"nominalPeakTflops":989.5,"mfuNominalPct":43.912569867194904,"configuredPeakTflops":989.5,"mfuConfiguredPct":43.912569867194904,"vramAllocatedGb":115.946832896,"vramAllocatedPct":77.23570906498168,"vramReservedGb":116.391936,"vramReservedPct":77.53220576942638,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.074281472,"vramAllocatedPct":99.30290931853295,"vramReservedGb":149.268987904,"vramReservedPct":99.4326091900984,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 65.44 MiB is free. Including non-PyTorch memory, this process has 139.74 GiB memory in use. Of the allocated memory 138.84 GiB is allocated by PyTorch, and 183.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":99.590923264,"vramAllocatedPct":97.66310200874165,"vramReservedGb":100.094967808,"vramReservedPct":98.15738956130434,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 94.97 GiB of which 1.50 GiB is free. Including non-PyTorch memory, this process has 93.47 GiB memory in use. Of the allocated memory 92.75 GiB is allocated by PyTorch, and 72.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"1b","modelLabel":"1B","parameters":995135232,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":99.590923264,"vramAllocatedPct":97.66310200874165,"vramReservedGb":100.094967808,"vramReservedPct":98.15738956130434,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 94.97 GiB of which 1.50 GiB is free. Including non-PyTorch memory, this process has 93.47 GiB memory in use. Of the allocated memory 92.75 GiB is allocated by PyTorch, and 72.69 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":9284.004382965368,"meanTps":9279.61525574384,"stepMs":441.1889343261719,"jitter":0.0014117993862990759,"achievedTflops":115.98829610447092,"nominalPeakTflops":312.0,"mfuNominalPct":37.17573593092017,"configuredPeakTflops":312.0,"mfuConfiguredPct":37.17573593092017,"vramAllocatedGb":45.6252032,"vramAllocatedPct":53.69298194070819,"vramReservedGb":46.045069312,"vramReservedPct":54.18709181831924,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10171.39168470486,"meanTps":10169.692973764677,"stepMs":805.3961791992188,"jitter":0.001452757132655699,"achievedTflops":127.07473433390143,"nominalPeakTflops":312.0,"mfuNominalPct":40.7290815172761,"configuredPeakTflops":312.0,"mfuConfiguredPct":40.7290815172761,"vramAllocatedGb":63.106288128,"vramAllocatedPct":74.26519886275995,"vramReservedGb":63.820529664,"vramReservedPct":75.10573775802018,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.268619776,"vramAllocatedPct":99.16960720081073,"vramReservedGb":84.387299328,"vramReservedPct":99.30927253039478,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 114.75 MiB is free. Process 764857 has 79.02 GiB memory in use. Of the allocated memory 78.40 GiB is allocated by PyTorch, and 113.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":9282.447664232237,"meanTps":9285.25948026634,"stepMs":441.26292419433594,"jitter":0.0024504067433123596,"achievedTflops":115.96884747584886,"nominalPeakTflops":312.0,"mfuNominalPct":37.1695023961054,"configuredPeakTflops":312.0,"mfuConfiguredPct":37.1695023961054,"vramAllocatedGb":45.6252032,"vramAllocatedPct":53.69298194070819,"vramReservedGb":46.045069312,"vramReservedPct":54.18709181831924,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10169.1279158462,"meanTps":10171.629101658371,"stepMs":805.5754699707031,"jitter":0.000805578424021399,"achievedTflops":127.04645228212082,"nominalPeakTflops":312.0,"mfuNominalPct":40.72001675709001,"configuredPeakTflops":312.0,"mfuConfiguredPct":40.72001675709001,"vramAllocatedGb":63.106288128,"vramAllocatedPct":74.26519886275995,"vramReservedGb":63.820529664,"vramReservedPct":75.10573775802018,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.268619776,"vramAllocatedPct":99.16960720081073,"vramReservedGb":84.387299328,"vramReservedPct":99.30927253039478,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 114.75 MiB is free. Process 852916 has 79.02 GiB memory in use. Of the allocated memory 78.40 GiB is allocated by PyTorch, and 113.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":41830.42139440059,"meanTps":41774.550997190345,"stepMs":391.6766662597656,"jitter":0.0014736519281955043,"achievedTflops":522.6020047740244,"nominalPeakTflops":2250.0,"mfuNominalPct":23.226755767734417,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.226755767734417,"vramAllocatedGb":98.068457984,"vramAllocatedPct":51.20987870597693,"vramReservedGb":99.37354752,"vramReservedPct":51.89137689828973,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":44888.31864322744,"meanTps":44877.15296254262,"stepMs":729.9894714355469,"jitter":0.001182594445428035,"achievedTflops":560.8053787625972,"nominalPeakTflops":2250.0,"mfuNominalPct":24.924683500559876,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.924683500559876,"vramAllocatedGb":167.992797184,"vramAllocatedPct":87.72332046532225,"vramReservedGb":170.597023744,"vramReservedPct":89.08320853741003,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.185536512,"vramAllocatedPct":99.31203627163852,"vramReservedGb":190.314446848,"vramReservedPct":99.37935131672248,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 640.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 599.81 MiB is free. Including non-PyTorch memory, this process has 177.75 GiB memory in use. Of the allocated memory 176.81 GiB is allocated by PyTorch, and 122.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":41854.74936814926,"meanTps":41770.348489276854,"stepMs":391.4490051269531,"jitter":0.001035750960025652,"achievedTflops":522.905942612309,"nominalPeakTflops":2250.0,"mfuNominalPct":23.240264116102622,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.240264116102622,"vramAllocatedGb":98.068457984,"vramAllocatedPct":51.20987870597693,"vramReservedGb":99.37354752,"vramReservedPct":51.89137689828973,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":44845.61834194118,"meanTps":44785.324713135764,"stepMs":730.6845397949219,"jitter":0.0019543164951563645,"achievedTflops":560.2719090457553,"nominalPeakTflops":2250.0,"mfuNominalPct":24.9009737353669,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.9009737353669,"vramAllocatedGb":167.992797184,"vramAllocatedPct":87.72332046532225,"vramReservedGb":170.597023744,"vramReservedPct":89.08320853741003,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.185536512,"vramAllocatedPct":99.31203627163852,"vramReservedGb":190.314446848,"vramReservedPct":99.37935131672248,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 640.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 599.81 MiB is free. Including non-PyTorch memory, this process has 177.75 GiB memory in use. Of the allocated memory 176.81 GiB is allocated by PyTorch, and 122.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":36354.20324827406,"meanTps":36334.45396918026,"stepMs":450.6769104003906,"jitter":0.0005349506537238058,"achievedTflops":454.18570662674574,"nominalPeakTflops":2250.0,"mfuNominalPct":20.186031405633145,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.186031405633145,"vramAllocatedGb":98.068457984,"vramAllocatedPct":34.11923659450369,"vramReservedGb":99.37354752,"vramReservedPct":34.573293480592994,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":42609.31148423638,"meanTps":42605.152022053735,"stepMs":769.0337829589844,"jitter":0.0007719613873165707,"achievedTflops":532.3329495954717,"nominalPeakTflops":2250.0,"mfuNominalPct":23.65924220424319,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.65924220424319,"vramAllocatedGb":167.992797184,"vramAllocatedPct":58.44678412531496,"vramReservedGb":170.597023744,"vramReservedPct":59.35282694451405,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.387175936,"vramAllocatedPct":98.59392410540532,"vramReservedGb":283.644002304,"vramReservedPct":98.68327719398889,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.79 GiB is free. Including non-PyTorch memory, this process has 264.89 GiB memory in use. Of the allocated memory 263.92 GiB is allocated by PyTorch, and 144.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":36355.215045062185,"meanTps":36321.86565866596,"stepMs":450.66436767578125,"jitter":0.00035977273542953147,"achievedTflops":454.19834735596316,"nominalPeakTflops":2250.0,"mfuNominalPct":20.186593215820587,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.186593215820587,"vramAllocatedGb":98.068457984,"vramAllocatedPct":34.11923659450369,"vramReservedGb":99.37354752,"vramReservedPct":34.573293480592994,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":42592.093593759215,"meanTps":42585.550997828,"stepMs":769.3446655273438,"jitter":0.0006581456417533978,"achievedTflops":532.1178405006694,"nominalPeakTflops":2250.0,"mfuNominalPct":23.64968180002975,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.64968180002975,"vramAllocatedGb":167.992797184,"vramAllocatedPct":58.44678412531496,"vramReservedGb":170.597023744,"vramReservedPct":59.35282694451405,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.387175936,"vramAllocatedPct":98.59392410540532,"vramReservedGb":283.644002304,"vramReservedPct":98.68327719398889,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.79 GiB is free. Including non-PyTorch memory, this process has 264.89 GiB memory in use. Of the allocated memory 263.92 GiB is allocated by PyTorch, and 144.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":20825.98621857143,"meanTps":20823.280126213678,"stepMs":196.67736053466797,"jitter":0.0009757513500869606,"achievedTflops":260.18628993965916,"nominalPeakTflops":989.5,"mfuNominalPct":26.29472359167854,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.29472359167854,"vramAllocatedGb":45.675272704,"vramAllocatedPct":53.72455811326761,"vramReservedGb":46.087012352,"vramReservedPct":54.20885802735604,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":24109.32620280127,"meanTps":24105.367382492685,"stepMs":339.7855224609375,"jitter":0.0007878993814672988,"achievedTflops":301.2061984396226,"nominalPeakTflops":989.5,"mfuNominalPct":30.440242389047256,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.440242389047256,"vramAllocatedGb":63.156357632,"vramAllocatedPct":74.2863086513231,"vramReservedGb":63.841501184,"vramReservedPct":75.09219403298017,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.972921344,"vramAllocatedPct":98.77134444106981,"vramReservedGb":84.093698048,"vramReservedPct":98.91340544407306,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 128.19 MiB is free. Including non-PyTorch memory, this process has 79.04 GiB memory in use. Of the allocated memory 78.21 GiB is allocated by PyTorch, and 115.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":20830.40054580236,"meanTps":20832.00723380493,"stepMs":196.63568115234375,"jitter":0.0006252321925927444,"achievedTflops":260.24143966523474,"nominalPeakTflops":989.5,"mfuNominalPct":26.300297085925692,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.300297085925692,"vramAllocatedGb":45.675272704,"vramAllocatedPct":53.72455811326761,"vramReservedGb":46.087012352,"vramReservedPct":54.20885802735604,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":24111.163651455212,"meanTps":24104.42601309031,"stepMs":339.75962829589844,"jitter":0.0011622967556579535,"achievedTflops":301.2291543247944,"nominalPeakTflops":989.5,"mfuNominalPct":30.442562337018135,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.442562337018135,"vramAllocatedGb":63.156357632,"vramAllocatedPct":74.2863086513231,"vramReservedGb":63.841501184,"vramReservedPct":75.09219403298017,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.972921344,"vramAllocatedPct":98.77134444106981,"vramReservedGb":84.093698048,"vramReservedPct":98.91340544407306,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 128.19 MiB is free. Including non-PyTorch memory, this process has 79.04 GiB memory in use. Of the allocated memory 78.21 GiB is allocated by PyTorch, and 115.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25831.912372866933,"meanTps":25872.70581928535,"stepMs":317.1271209716797,"jitter":0.005181656019707799,"achievedTflops":322.7270666466264,"nominalPeakTflops":989.5,"mfuNominalPct":32.615165906682805,"configuredPeakTflops":989.5,"mfuConfiguredPct":32.615165906682805,"vramAllocatedGb":63.156357632,"vramAllocatedPct":42.070369166912954,"vramReservedGb":63.856181248,"vramReservedPct":42.53654294546421,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":27719.62718927991,"meanTps":27746.00379565638,"stepMs":591.0613403320312,"jitter":0.0023817564714560145,"achievedTflops":346.31094447079545,"nominalPeakTflops":989.5,"mfuNominalPct":34.9985795321673,"configuredPeakTflops":989.5,"mfuConfiguredPct":34.9985795321673,"vramAllocatedGb":98.118527488,"vramAllocatedPct":65.35973302302261,"vramReservedGb":99.46791936,"vramReservedPct":66.25860395754762,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.311973888,"vramAllocatedPct":98.79511306998286,"vramReservedGb":148.497235968,"vramReservedPct":98.9185217716633,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 819.44 MiB is free. Including non-PyTorch memory, this process has 139.00 GiB memory in use. Of the allocated memory 138.13 GiB is allocated by PyTorch, and 156.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25844.937283036266,"meanTps":25880.430109293404,"stepMs":316.96730041503906,"jitter":0.0057454689839375765,"achievedTflops":322.88979137995665,"nominalPeakTflops":989.5,"mfuNominalPct":32.63161105406333,"configuredPeakTflops":989.5,"mfuConfiguredPct":32.63161105406333,"vramAllocatedGb":63.156357632,"vramAllocatedPct":42.070369166912954,"vramReservedGb":63.856181248,"vramReservedPct":42.53654294546421,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":27770.8690714918,"meanTps":27782.448731877517,"stepMs":589.9707336425781,"jitter":0.0040653056531031375,"achievedTflops":346.95112712924487,"nominalPeakTflops":989.5,"mfuNominalPct":35.06327712271297,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.06327712271297,"vramAllocatedGb":98.118527488,"vramAllocatedPct":65.35973302302261,"vramReservedGb":99.46791936,"vramReservedPct":66.25860395754762,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.311973888,"vramAllocatedPct":98.79511306998286,"vramReservedGb":148.497235968,"vramReservedPct":98.9185217716633,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 819.44 MiB is free. Including non-PyTorch memory, this process has 139.00 GiB memory in use. Of the allocated memory 138.13 GiB is allocated by PyTorch, and 156.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":5263.9150982245455,"meanTps":5264.916885474438,"stepMs":389.06402587890625,"jitter":0.0030575714044796193,"achievedTflops":65.76392232234701,"nominalPeakTflops":364.2,"mfuNominalPct":18.05709014891461,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.567797366515654,"vramAllocatedGb":36.884660736,"vramAllocatedPct":72.49793677760044,"vramReservedGb":37.102813184,"vramReservedPct":72.92672213355593,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":5905.209704489012,"meanTps":5905.464332142805,"stepMs":693.6248168945312,"jitter":0.0024124125279855447,"achievedTflops":73.77583890632485,"nominalPeakTflops":364.2,"mfuNominalPct":20.256957415245704,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.342569508918032,"vramAllocatedGb":45.6252032,"vramAllocatedPct":89.67774221196441,"vramReservedGb":46.049263616,"vramReservedPct":90.51124602128765,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":50.208507904,"vramAllocatedPct":98.686368780102,"vramReservedGb":50.2792192,"vramReservedPct":98.82535401040036,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 68.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 44.50 MiB is free. Process 2319972 has 47.33 GiB memory in use. Of the allocated memory 46.76 GiB is allocated by PyTorch, and 67.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":3747.4611791127463,"meanTps":3747.900852376773,"stepMs":546.5033264160156,"jitter":0.0016900881687392341,"achievedTflops":46.818336027552085,"nominalPeakTflops":154.8,"mfuNominalPct":30.244403118573697,"configuredPeakTflops":180.6,"mfuConfiguredPct":25.9237741016346,"vramAllocatedGb":36.884660736,"vramAllocatedPct":72.27477768692309,"vramReservedGb":37.081841664,"vramReservedPct":72.66114989290062,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":4574.182416656931,"meanTps":4574.593175806858,"stepMs":895.4605712890625,"jitter":0.0008962058181095431,"achievedTflops":57.14685201490684,"nominalPeakTflops":154.8,"mfuNominalPct":36.91657106906127,"configuredPeakTflops":180.6,"mfuConfiguredPct":31.64277520205252,"vramAllocatedGb":45.6252032,"vramAllocatedPct":89.40170120589535,"vramReservedGb":46.049263616,"vramReservedPct":90.23263937045084,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.54772224,"vramAllocatedPct":99.04728183959236,"vramReservedGb":50.61476352,"vramReservedPct":99.17864792812784,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 108.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 62.25 MiB is free. Process 1415479 has 47.46 GiB memory in use. Of the allocated memory 47.08 GiB is allocated by PyTorch, and 63.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":3738.6959914049203,"meanTps":3738.585561343724,"stepMs":547.7845764160156,"jitter":0.0006658143974266015,"achievedTflops":46.70882948863532,"nominalPeakTflops":154.8,"mfuNominalPct":30.17366246035873,"configuredPeakTflops":180.6,"mfuConfiguredPct":25.863139251736058,"vramAllocatedGb":36.884660736,"vramAllocatedPct":72.27477768692309,"vramReservedGb":37.081841664,"vramReservedPct":72.66114989290062,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":4573.7542280766265,"meanTps":4574.35817331846,"stepMs":895.5444030761719,"jitter":0.0006485604749178902,"achievedTflops":57.14150250603199,"nominalPeakTflops":154.8,"mfuNominalPct":36.91311531397415,"configuredPeakTflops":180.6,"mfuConfiguredPct":31.63981312626356,"vramAllocatedGb":45.6252032,"vramAllocatedPct":89.40170120589535,"vramReservedGb":46.049263616,"vramReservedPct":90.23263937045084,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.54772224,"vramAllocatedPct":99.04728183959236,"vramReservedGb":50.61476352,"vramReservedPct":99.17864792812784,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 108.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 62.25 MiB is free. Process 1449495 has 47.46 GiB memory in use. Of the allocated memory 47.08 GiB is allocated by PyTorch, and 63.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14478.56780786652,"meanTps":14475.819382914235,"stepMs":565.8018188476562,"jitter":0.0003858398137581516,"achievedTflops":180.88578384870272,"nominalPeakTflops":468.0,"mfuNominalPct":38.650808514680065,"configuredPeakTflops":468.0,"mfuConfiguredPct":38.650808514680065,"vramAllocatedGb":63.106288128,"vramAllocatedPct":61.88471451861473,"vramReservedGb":63.814238208,"vramReservedPct":62.578960526324245,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":15039.481885448684,"meanTps":15039.57491001006,"stepMs":1089.3992309570312,"jitter":0.00015832149728184944,"achievedTflops":187.89347852828945,"nominalPeakTflops":468.0,"mfuNominalPct":40.14817917271142,"configuredPeakTflops":468.0,"mfuConfiguredPct":40.14817917271142,"vramAllocatedGb":98.068457984,"vramAllocatedPct":96.17010769688798,"vramReservedGb":99.42597632,"vramReservedPct":97.50134800857842,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.816320512,"vramAllocatedPct":98.86477875307118,"vramReservedGb":101.122572288,"vramReservedPct":99.16510229119685,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 161.75 MiB is free. Including non-PyTorch memory, this process has 94.80 GiB memory in use. Of the allocated memory 93.89 GiB is allocated by PyTorch, and 272.06 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14460.035913683263,"meanTps":14455.45911921683,"stepMs":566.5269470214844,"jitter":0.0002765705799222202,"achievedTflops":180.6542584485372,"nominalPeakTflops":468.0,"mfuNominalPct":38.60133727532847,"configuredPeakTflops":468.0,"mfuConfiguredPct":38.60133727532847,"vramAllocatedGb":63.272462848,"vramAllocatedPct":62.04767252822785,"vramReservedGb":69.28990208,"vramReservedPct":67.94862978703713,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":98.237016576,"vramAllocatedPct":96.33540343293923,"vramReservedGb":100.787027968,"vramReservedPct":98.83605323653806,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 94.97 GiB of which 1.43 GiB is free. Including non-PyTorch memory, this process has 93.54 GiB memory in use. Of the allocated memory 89.54 GiB is allocated by PyTorch, and 3.35 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":9138.007110704486,"meanTps":9140.45432838195,"stepMs":448.23777770996094,"jitter":0.0013786963045762533,"achievedTflops":118.18869727495245,"nominalPeakTflops":312.0,"mfuNominalPct":37.880992716330915,"configuredPeakTflops":312.0,"mfuConfiguredPct":37.880992716330915,"vramAllocatedGb":45.630153216,"vramAllocatedPct":53.69880725437373,"vramReservedGb":46.049263616,"vramReservedPct":54.19202778951785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10019.86298639375,"meanTps":10021.330743402918,"stepMs":817.5760498046875,"jitter":0.0012467742243731806,"achievedTflops":129.59440049550304,"nominalPeakTflops":312.0,"mfuNominalPct":41.53666682548174,"configuredPeakTflops":312.0,"mfuConfiguredPct":41.53666682548174,"vramAllocatedGb":63.112518144,"vramAllocatedPct":74.27253051529226,"vramReservedGb":63.824723968,"vramReservedPct":75.11067372921879,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.272289792,"vramAllocatedPct":99.17392617560951,"vramReservedGb":84.391493632,"vramReservedPct":99.3142085015934,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 110.75 MiB is free. Process 764857 has 79.02 GiB memory in use. Of the allocated memory 78.41 GiB is allocated by PyTorch, and 113.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":9141.995258233384,"meanTps":9139.758728348357,"stepMs":448.042236328125,"jitter":0.0010258519527501647,"achievedTflops":118.24027897709718,"nominalPeakTflops":312.0,"mfuNominalPct":37.89752531317217,"configuredPeakTflops":312.0,"mfuConfiguredPct":37.89752531317217,"vramAllocatedGb":45.630153216,"vramAllocatedPct":53.69880725437373,"vramReservedGb":46.049263616,"vramReservedPct":54.19202778951785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10031.661182120708,"meanTps":10032.265881401967,"stepMs":816.614501953125,"jitter":0.0011917835402632043,"achievedTflops":129.74699540665503,"nominalPeakTflops":312.0,"mfuNominalPct":41.58557545085097,"configuredPeakTflops":312.0,"mfuConfiguredPct":41.58557545085097,"vramAllocatedGb":63.112518144,"vramAllocatedPct":74.27253051529226,"vramReservedGb":63.824723968,"vramReservedPct":75.11067372921879,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.272289792,"vramAllocatedPct":99.17392617560951,"vramReservedGb":84.391493632,"vramReservedPct":99.3142085015934,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 110.75 MiB is free. Process 852916 has 79.02 GiB memory in use. Of the allocated memory 78.41 GiB is allocated by PyTorch, and 113.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":40574.711735025296,"meanTps":40561.81068455451,"stepMs":403.7983093261719,"jitter":0.000711569762621675,"achievedTflops":524.7831681649526,"nominalPeakTflops":2250.0,"mfuNominalPct":23.323696362886782,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.323696362886782,"vramAllocatedGb":98.077248,"vramAllocatedPct":51.21446872056915,"vramReservedGb":99.377741824,"vramReservedPct":51.893567100965605,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":43590.79592466901,"meanTps":43564.775526763115,"stepMs":751.7183227539062,"jitter":0.0023842767217977885,"achievedTflops":563.7924463288967,"nominalPeakTflops":2250.0,"mfuNominalPct":25.05744205906208,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.05744205906208,"vramAllocatedGb":168.006707712,"vramAllocatedPct":87.73058433452402,"vramReservedGb":170.622189568,"vramReservedPct":89.0963497534653,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.189206528,"vramAllocatedPct":99.31395269897992,"vramReservedGb":190.318641152,"vramReservedPct":99.38154151939835,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 640.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 595.81 MiB is free. Including non-PyTorch memory, this process has 177.76 GiB memory in use. Of the allocated memory 176.82 GiB is allocated by PyTorch, and 123.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":40607.08744388285,"meanTps":40496.68996110583,"stepMs":403.4763641357422,"jitter":0.0007163427497994023,"achievedTflops":525.2019074816188,"nominalPeakTflops":2250.0,"mfuNominalPct":23.34230699918306,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.34230699918306,"vramAllocatedGb":98.077248,"vramAllocatedPct":51.21446872056915,"vramReservedGb":99.377741824,"vramReservedPct":51.893567100965605,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":43626.93222056011,"meanTps":43606.87230696577,"stepMs":751.0956726074219,"jitter":0.0019101230571240263,"achievedTflops":564.2598241372057,"nominalPeakTflops":2250.0,"mfuNominalPct":25.078214406098027,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.078214406098027,"vramAllocatedGb":168.006707712,"vramAllocatedPct":87.73058433452402,"vramReservedGb":170.622189568,"vramReservedPct":89.0963497534653,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.189206528,"vramAllocatedPct":99.31395269897992,"vramReservedGb":190.318641152,"vramReservedPct":99.38154151939835,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 640.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 595.81 MiB is free. Including non-PyTorch memory, this process has 177.76 GiB memory in use. Of the allocated memory 176.82 GiB is allocated by PyTorch, and 123.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":35454.07812503869,"meanTps":35374.659653648625,"stepMs":462.11891174316406,"jitter":0.0005611135125229381,"achievedTflops":458.5541744407405,"nominalPeakTflops":2250.0,"mfuNominalPct":20.380185530699578,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.380185530699578,"vramAllocatedGb":98.077248,"vramAllocatedPct":34.12229475042598,"vramReservedGb":99.377741824,"vramReservedPct":34.5747527311356,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":41436.59257753931,"meanTps":41438.72840908103,"stepMs":790.7986145019531,"jitter":0.0003011213377250217,"achievedTflops":535.9305193049668,"nominalPeakTflops":2250.0,"mfuNominalPct":23.819134191331855,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.819134191331855,"vramAllocatedGb":168.006707712,"vramAllocatedPct":58.45162377106593,"vramReservedGb":170.622189568,"vramReservedPct":59.36158244776966,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.390845952,"vramAllocatedPct":98.5952009496301,"vramReservedGb":283.648196608,"vramReservedPct":98.6847364445315,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.78 GiB is free. Including non-PyTorch memory, this process has 264.89 GiB memory in use. Of the allocated memory 263.93 GiB is allocated by PyTorch, and 145.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":35436.21108360716,"meanTps":35436.76519430822,"stepMs":462.35191345214844,"jitter":0.0004008799732151358,"achievedTflops":458.32308659791346,"nominalPeakTflops":2250.0,"mfuNominalPct":20.369914959907263,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.369914959907263,"vramAllocatedGb":98.077248,"vramAllocatedPct":34.12229475042598,"vramReservedGb":99.377741824,"vramReservedPct":34.5747527311356,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":41417.87429699976,"meanTps":41420.87949408054,"stepMs":791.156005859375,"jitter":0.0004252568448500651,"achievedTflops":535.6884217485309,"nominalPeakTflops":2250.0,"mfuNominalPct":23.808374299934705,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.808374299934705,"vramAllocatedGb":168.006707712,"vramAllocatedPct":58.45162377106593,"vramReservedGb":170.622189568,"vramReservedPct":59.36158244776966,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.390845952,"vramAllocatedPct":98.5952009496301,"vramReservedGb":283.648196608,"vramReservedPct":98.6847364445315,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.78 GiB is free. Including non-PyTorch memory, this process has 264.89 GiB memory in use. Of the allocated memory 263.93 GiB is allocated by PyTorch, and 145.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":20458.489023858467,"meanTps":20454.15094813917,"stepMs":200.21028900146484,"jitter":0.0008229961379032506,"achievedTflops":264.6049775022921,"nominalPeakTflops":989.5,"mfuNominalPct":26.741281202859234,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.741281202859234,"vramAllocatedGb":45.68022272,"vramAllocatedPct":53.73038046322876,"vramReservedGb":46.091206656,"vramReservedPct":54.21379148731484,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":23662.697018382714,"meanTps":23661.591451159584,"stepMs":346.19891357421875,"jitter":0.0004983932030387916,"achievedTflops":306.0474018824601,"nominalPeakTflops":989.5,"mfuNominalPct":30.929499937590712,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.929499937590712,"vramAllocatedGb":63.162587648,"vramAllocatedPct":74.29363657378144,"vramReservedGb":63.866667008,"vramReservedPct":75.12179479273301,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.97659136,"vramAllocatedPct":98.77566121853377,"vramReservedGb":84.097892352,"vramReservedPct":98.91833890403187,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 124.19 MiB is free. Including non-PyTorch memory, this process has 79.05 GiB memory in use. Of the allocated memory 78.21 GiB is allocated by PyTorch, and 115.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":20467.436617942072,"meanTps":20462.055580301836,"stepMs":200.12276458740234,"jitter":0.0005485460794462444,"achievedTflops":264.7207034451235,"nominalPeakTflops":989.5,"mfuNominalPct":26.752976598799748,"configuredPeakTflops":989.5,"mfuConfiguredPct":26.752976598799748,"vramAllocatedGb":45.68022272,"vramAllocatedPct":53.73038046322876,"vramReservedGb":46.091206656,"vramReservedPct":54.21379148731484,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":23653.086075507792,"meanTps":23650.783486434015,"stepMs":346.33958435058594,"jitter":0.0002633676673387902,"achievedTflops":305.92309635235824,"nominalPeakTflops":989.5,"mfuNominalPct":30.916937478762836,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.916937478762836,"vramAllocatedGb":63.162587648,"vramAllocatedPct":74.29363657378144,"vramReservedGb":63.866667008,"vramReservedPct":75.12179479273301,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.97659136,"vramAllocatedPct":98.77566121853377,"vramReservedGb":84.097892352,"vramReservedPct":98.91833890403187,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 124.19 MiB is free. Including non-PyTorch memory, this process has 79.05 GiB memory in use. Of the allocated memory 78.21 GiB is allocated by PyTorch, and 115.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25282.649196315266,"meanTps":25295.69935346975,"stepMs":324.0166778564453,"jitter":0.0054788753739918004,"achievedTflops":326.99945797500675,"nominalPeakTflops":989.5,"mfuNominalPct":33.04693865336096,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.04693865336096,"vramAllocatedGb":63.162587648,"vramAllocatedPct":42.07451916990336,"vramReservedGb":63.860375552,"vramReservedPct":42.539336898825276,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":27144.437619344793,"meanTps":27161.121169634902,"stepMs":603.5859069824219,"jitter":0.0020127759494836337,"achievedTflops":351.0793635445357,"nominalPeakTflops":989.5,"mfuNominalPct":35.4804814092507,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.4804814092507,"vramAllocatedGb":98.127317504,"vramAllocatedPct":65.36558831981249,"vramReservedGb":99.472113664,"vramReservedPct":66.26139791090868,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.315643904,"vramAllocatedPct":98.79755777917379,"vramReservedGb":148.503527424,"vramReservedPct":98.92271270170488,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 813.44 MiB is free. Including non-PyTorch memory, this process has 139.01 GiB memory in use. Of the allocated memory 138.13 GiB is allocated by PyTorch, and 159.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":25274.829204408932,"meanTps":25271.902534477034,"stepMs":324.11692810058594,"jitter":0.0024674953311532746,"achievedTflops":326.898316156565,"nominalPeakTflops":989.5,"mfuNominalPct":33.036717145686204,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.036717145686204,"vramAllocatedGb":63.162587648,"vramAllocatedPct":42.07451916990336,"vramReservedGb":63.860375552,"vramReservedPct":42.539336898825276,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":27101.618047697393,"meanTps":27100.05329545265,"stepMs":604.53955078125,"jitter":0.0021798679546109873,"achievedTflops":350.5255459200178,"nominalPeakTflops":989.5,"mfuNominalPct":35.424511967662234,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.424511967662234,"vramAllocatedGb":98.127317504,"vramAllocatedPct":65.36558831981249,"vramReservedGb":99.472113664,"vramReservedPct":66.26139791090868,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.315643904,"vramAllocatedPct":98.79755777917379,"vramReservedGb":148.503527424,"vramReservedPct":98.92271270170488,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 813.44 MiB is free. Including non-PyTorch memory, this process has 139.01 GiB memory in use. Of the allocated memory 138.13 GiB is allocated by PyTorch, and 159.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":5219.588808284158,"meanTps":5218.812386607891,"stepMs":392.3680725097656,"jitter":0.001937267896060991,"achievedTflops":67.50885549644393,"nominalPeakTflops":364.2,"mfuNominalPct":18.53620414509718,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":14.954328947976952,"vramAllocatedGb":36.888970752,"vramAllocatedPct":72.50640824680318,"vramReservedGb":37.104910336,"vramReservedPct":72.93084415040724,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":5835.781747981443,"meanTps":5832.580930548807,"stepMs":701.8768310546875,"jitter":0.003583455500432625,"achievedTflops":75.47854078236738,"nominalPeakTflops":364.2,"mfuNominalPct":20.7244757776956,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":16.719746158817223,"vramAllocatedGb":45.630153216,"vramAllocatedPct":89.68747162088007,"vramReservedGb":46.051360768,"vramReservedPct":90.51536803813896,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":50.21217792,"vramAllocatedPct":98.6935823095918,"vramReservedGb":50.283413504,"vramReservedPct":98.833598044103,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 68.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 40.50 MiB is free. Process 2319972 has 47.33 GiB memory in use. Of the allocated memory 46.76 GiB is allocated by PyTorch, and 67.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":3714.583015638211,"meanTps":3713.73686374776,"stepMs":551.3404846191406,"jitter":0.002288198391205738,"achievedTflops":48.04348718701078,"nominalPeakTflops":154.8,"mfuNominalPct":31.035844436053473,"configuredPeakTflops":180.6,"mfuConfiguredPct":26.60215237376012,"vramAllocatedGb":36.888970752,"vramAllocatedPct":72.28322307972354,"vramReservedGb":37.083938816,"vramReservedPct":72.66525922159042,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":4526.275317868984,"meanTps":4525.594441682896,"stepMs":904.9383239746094,"jitter":0.0008596204588814379,"achievedTflops":58.541712306181886,"nominalPeakTflops":154.8,"mfuNominalPct":37.81764360864463,"configuredPeakTflops":180.6,"mfuConfiguredPct":32.415123093123974,"vramAllocatedGb":45.630153216,"vramAllocatedPct":89.41140066628913,"vramReservedGb":46.051360768,"vramReservedPct":90.23674869914063,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.551392256,"vramAllocatedPct":99.05447316479949,"vramReservedGb":50.618957824,"vramReservedPct":99.18686658550743,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 108.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 58.25 MiB is free. Process 1415479 has 47.46 GiB memory in use. Of the allocated memory 47.08 GiB is allocated by PyTorch, and 64.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":3702.6581132400274,"meanTps":3701.6574643537065,"stepMs":553.1161499023438,"jitter":0.0014994092529949773,"achievedTflops":47.88925348348026,"nominalPeakTflops":154.8,"mfuNominalPct":30.93621026064616,"configuredPeakTflops":180.6,"mfuConfiguredPct":26.51675165198243,"vramAllocatedGb":36.888970752,"vramAllocatedPct":72.28322307972354,"vramReservedGb":37.083938816,"vramReservedPct":72.66525922159042,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":4525.68650434771,"meanTps":4525.573353099811,"stepMs":905.0560607910156,"jitter":0.0005488097347539324,"achievedTflops":58.53409673944243,"nominalPeakTflops":154.8,"mfuNominalPct":37.81272399188787,"configuredPeakTflops":180.6,"mfuConfiguredPct":32.41090627876104,"vramAllocatedGb":45.630153216,"vramAllocatedPct":89.41140066628913,"vramReservedGb":46.051360768,"vramReservedPct":90.23674869914063,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.551392256,"vramAllocatedPct":99.05447316479949,"vramReservedGb":50.618957824,"vramReservedPct":99.18686658550743,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 108.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 58.25 MiB is free. Process 1449495 has 47.46 GiB memory in use. Of the allocated memory 47.08 GiB is allocated by PyTorch, and 64.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14372.386902809558,"meanTps":14372.152896660236,"stepMs":569.9818725585938,"jitter":0.00021289810242089188,"achievedTflops":185.88885565484028,"nominalPeakTflops":468.0,"mfuNominalPct":39.719840951888955,"configuredPeakTflops":468.0,"mfuConfiguredPct":39.719840951888955,"vramAllocatedGb":63.112518144,"vramAllocatedPct":61.890823937708184,"vramReservedGb":63.818432512,"vramReservedPct":62.58307363950748,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":14941.95039727313,"meanTps":14942.511327143022,"stepMs":1096.5101318359375,"jitter":0.0003110409323079893,"achievedTflops":193.25544736466324,"nominalPeakTflops":468.0,"mfuNominalPct":41.29389900954342,"configuredPeakTflops":468.0,"mfuConfiguredPct":41.29389900954342,"vramAllocatedGb":98.077248,"vramAllocatedPct":96.17872756103957,"vramReservedGb":99.430170624,"vramReservedPct":97.50546112176164,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.819990528,"vramAllocatedPct":98.8683777271065,"vramReservedGb":101.128863744,"vramReservedPct":99.1712719609717,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 155.75 MiB is free. Including non-PyTorch memory, this process has 94.81 GiB memory in use. Of the allocated memory 93.90 GiB is allocated by PyTorch, and 274.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14353.196196071274,"meanTps":14347.243461129365,"stepMs":570.7439575195312,"jitter":0.0007500939839070806,"achievedTflops":185.64064785616975,"nominalPeakTflops":468.0,"mfuNominalPct":39.66680509747217,"configuredPeakTflops":468.0,"mfuConfiguredPct":39.66680509747217,"vramAllocatedGb":63.280228864,"vramAllocatedPct":62.05528821435618,"vramReservedGb":69.291999232,"vramReservedPct":67.95068634362876,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":98.246830592,"vramAllocatedPct":96.34502747511407,"vramReservedGb":100.807999488,"vramReservedPct":98.85661880245425,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 94.97 GiB of which 1.41 GiB is free. Including non-PyTorch memory, this process has 93.55 GiB memory in use. Of the allocated memory 89.55 GiB is allocated by PyTorch, and 3.36 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":7433.793037748932,"meanTps":7427.684509100298,"stepMs":275.4986572265625,"jitter":0.0029780789783263213,"achievedTflops":102.69453237128944,"nominalPeakTflops":312.0,"mfuNominalPct":32.91491422156713,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.91491422156713,"vramAllocatedGb":36.8966144,"vramAllocatedPct":43.42094087709123,"vramReservedGb":36.987469824,"vramReservedPct":43.52786201492514,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":8858.294132419285,"meanTps":8862.681151239245,"stepMs":462.3915100097656,"jitter":0.003287785583677548,"achievedTflops":122.3733790968717,"nominalPeakTflops":312.0,"mfuNominalPct":39.222236890022984,"configuredPeakTflops":312.0,"mfuConfiguredPct":39.222236890022984,"vramAllocatedGb":45.638133248,"vramAllocatedPct":53.70819837340468,"vramReservedGb":46.080720896,"vramReservedPct":54.22904757350741,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":9716.026053463225,"meanTps":9719.499736802542,"stepMs":843.14306640625,"jitter":0.0018308114647863016,"achievedTflops":134.2225626945642,"nominalPeakTflops":312.0,"mfuNominalPct":43.020052145693654,"configuredPeakTflops":312.0,"mfuConfiguredPct":43.020052145693654,"vramAllocatedGb":63.121138176,"vramAllocatedPct":74.28267480375658,"vramReservedGb":63.810043904,"vramReservedPct":75.09339783002366,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.279629824,"vramAllocatedPct":99.18256412520708,"vramReservedGb":84.376813568,"vramReservedPct":99.29693260239827,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 124.75 MiB is free. Process 703773 has 79.01 GiB memory in use. Of the allocated memory 78.41 GiB is allocated by PyTorch, and 92.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":7408.794877456392,"meanTps":7402.463393117887,"stepMs":276.42822265625,"jitter":0.004996286707882535,"achievedTflops":102.34919394602676,"nominalPeakTflops":312.0,"mfuNominalPct":32.804228828854725,"configuredPeakTflops":312.0,"mfuConfiguredPct":32.804228828854725,"vramAllocatedGb":36.8966144,"vramAllocatedPct":43.42094087709123,"vramReservedGb":36.987469824,"vramReservedPct":43.52786201492514,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":8864.805400291509,"meanTps":8859.630212564654,"stepMs":462.0518798828125,"jitter":0.0018466260198793751,"achievedTflops":122.46332935589652,"nominalPeakTflops":312.0,"mfuNominalPct":39.251067101248886,"configuredPeakTflops":312.0,"mfuConfiguredPct":39.251067101248886,"vramAllocatedGb":45.638133248,"vramAllocatedPct":53.70819837340468,"vramReservedGb":46.080720896,"vramReservedPct":54.22904757350741,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":9720.439427499736,"meanTps":9722.345411602228,"stepMs":842.76025390625,"jitter":0.0017513271716880854,"achievedTflops":134.28353148674842,"nominalPeakTflops":312.0,"mfuNominalPct":43.03959342523988,"configuredPeakTflops":312.0,"mfuConfiguredPct":43.03959342523988,"vramAllocatedGb":63.121138176,"vramAllocatedPct":74.28267480375658,"vramReservedGb":63.810043904,"vramReservedPct":75.09339783002366,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.279629824,"vramAllocatedPct":99.18256412520708,"vramReservedGb":84.376813568,"vramReservedPct":99.29693260239827,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 124.75 MiB is free. Process 736416 has 79.01 GiB memory in use. Of the allocated memory 78.41 GiB is allocated by PyTorch, and 92.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11367.869843055609,"meanTps":11399.12267479854,"stepMs":180.1568832397461,"jitter":0.023691879089791045,"achievedTflops":157.04204726471528,"nominalPeakTflops":2250.0,"mfuNominalPct":6.979646545098457,"configuredPeakTflops":2250.0,"mfuConfiguredPct":6.979646545098457,"vramAllocatedGb":36.8966144,"vramAllocatedPct":19.2668589567654,"vramReservedGb":36.987469824,"vramReservedPct":19.314302297248833,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":22723.829018153803,"meanTps":22856.747283775007,"stepMs":180.25131225585938,"jitter":0.02494927781798788,"achievedTflops":313.91955396852114,"nominalPeakTflops":2250.0,"mfuNominalPct":13.951980176378717,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.951980176378717,"vramAllocatedGb":45.638133248,"vramAllocatedPct":23.83154906319214,"vramReservedGb":46.080720896,"vramReservedPct":24.06266169855693,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":33886.253607946994,"meanTps":33813.264509167355,"stepMs":241.74994659423828,"jitter":0.003083376137375713,"achievedTflops":468.1234667701771,"nominalPeakTflops":2250.0,"mfuNominalPct":20.80548741200787,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.80548741200787,"vramAllocatedGb":63.121138176,"vramAllocatedPct":32.96091216508721,"vramReservedGb":63.810043904,"vramReservedPct":33.3206484095022,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":38227.03338837712,"meanTps":38225.80585329388,"stepMs":428.5972137451172,"jitter":0.0006140783430342623,"achievedTflops":528.0894017127254,"nominalPeakTflops":2250.0,"mfuNominalPct":23.47064007612113,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.47064007612113,"vramAllocatedGb":98.087148032,"vramAllocatedPct":51.219638368877355,"vramReservedGb":99.413393408,"vramReservedPct":51.912183823710585,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":41075.87375121698,"meanTps":41071.91306658997,"stepMs":797.7432250976562,"jitter":0.0011546385314713404,"achievedTflops":567.4448596030202,"nominalPeakTflops":2250.0,"mfuNominalPct":25.219771537912006,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.219771537912006,"vramAllocatedGb":168.019167744,"vramAllocatedPct":87.73709077645765,"vramReservedGb":170.653646848,"vramReservedPct":89.11277627353441,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.19654656,"vramAllocatedPct":99.3177855536627,"vramReservedGb":190.366875648,"vramReservedPct":99.40672885017098,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 640.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 549.81 MiB is free. Including non-PyTorch memory, this process has 177.80 GiB memory in use. Of the allocated memory 176.82 GiB is allocated by PyTorch, and 162.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":12079.898967062052,"meanTps":12017.408310164192,"stepMs":169.537841796875,"jitter":0.015237011062717895,"achievedTflops":166.87841176306335,"nominalPeakTflops":2250.0,"mfuNominalPct":7.4168183005805925,"configuredPeakTflops":2250.0,"mfuConfiguredPct":7.4168183005805925,"vramAllocatedGb":36.8966144,"vramAllocatedPct":19.2668589567654,"vramReservedGb":36.987469824,"vramReservedPct":19.314302297248833,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":23945.82535504765,"meanTps":23906.935823267468,"stepMs":171.0527801513672,"jitter":0.015836610998192256,"achievedTflops":330.8008878635448,"nominalPeakTflops":2250.0,"mfuNominalPct":14.702261682824211,"configuredPeakTflops":2250.0,"mfuConfiguredPct":14.702261682824211,"vramAllocatedGb":45.638133248,"vramAllocatedPct":23.83154906319214,"vramReservedGb":46.080720896,"vramReservedPct":24.06266169855693,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":34239.406302828334,"meanTps":34179.47407175648,"stepMs":239.25648498535156,"jitter":0.0024339169866567497,"achievedTflops":473.0021136025999,"nominalPeakTflops":2250.0,"mfuNominalPct":21.022316160115547,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.022316160115547,"vramAllocatedGb":63.121138176,"vramAllocatedPct":32.96091216508721,"vramReservedGb":63.810043904,"vramReservedPct":33.3206484095022,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":38352.11773345063,"meanTps":38349.398326663446,"stepMs":427.19935607910156,"jitter":0.000927414802555008,"achievedTflops":529.8173861022642,"nominalPeakTflops":2250.0,"mfuNominalPct":23.547439382322853,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.547439382322853,"vramAllocatedGb":98.087148032,"vramAllocatedPct":51.219638368877355,"vramReservedGb":99.413393408,"vramReservedPct":51.912183823710585,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":41060.31068484799,"meanTps":41050.567626912285,"stepMs":798.0455932617188,"jitter":0.0012498458720909288,"achievedTflops":567.2298627884852,"nominalPeakTflops":2250.0,"mfuNominalPct":25.210216123932675,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.210216123932675,"vramAllocatedGb":168.019167744,"vramAllocatedPct":87.73709077645765,"vramReservedGb":170.653646848,"vramReservedPct":89.11277627353441,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.19654656,"vramAllocatedPct":99.3177855536627,"vramReservedGb":190.366875648,"vramReservedPct":99.40672885017098,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 640.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 549.81 MiB is free. Including non-PyTorch memory, this process has 177.80 GiB memory in use. Of the allocated memory 176.82 GiB is allocated by PyTorch, and 162.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11864.47869906821,"meanTps":11846.989558999516,"stepMs":172.61609649658203,"jitter":0.0005835347206958349,"achievedTflops":163.90247692433607,"nominalPeakTflops":2250.0,"mfuNominalPct":7.284554529970492,"configuredPeakTflops":2250.0,"mfuConfiguredPct":7.284554529970492,"vramAllocatedGb":36.8966144,"vramAllocatedPct":12.836791177599228,"vramReservedGb":36.987469824,"vramReservedPct":12.868400909933916,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":19001.149084991484,"meanTps":18996.74512111748,"stepMs":215.56591033935547,"jitter":0.0005024792410312566,"achievedTflops":262.4923924962058,"nominalPeakTflops":2250.0,"mfuNominalPct":11.666328555386924,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.666328555386924,"vramAllocatedGb":45.638133248,"vramAllocatedPct":15.878074337357749,"vramReservedGb":46.080720896,"vramReservedPct":16.032056086294606,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":26796.091062532807,"meanTps":26787.88439441107,"stepMs":305.7162322998047,"jitter":0.0007207497881626533,"achievedTflops":370.1760362538404,"nominalPeakTflops":2250.0,"mfuNominalPct":16.45226827794846,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.45226827794846,"vramAllocatedGb":63.121138176,"vramAllocatedPct":21.960629256479926,"vramReservedGb":63.810043904,"vramReservedPct":22.200308129872386,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":33620.767876317965,"meanTps":33592.51499443632,"stepMs":487.3178405761719,"jitter":0.0005388299639457377,"achievedTflops":464.45589990055277,"nominalPeakTflops":2250.0,"mfuNominalPct":20.642484440024568,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.642484440024568,"vramAllocatedGb":98.087148032,"vramAllocatedPct":34.12573909472429,"vramReservedGb":99.413393408,"vramReservedPct":34.58715636074771,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":39090.648801221294,"meanTps":39085.99414091117,"stepMs":838.2567443847656,"jitter":0.0003763714478343652,"achievedTflops":540.0198631232472,"nominalPeakTflops":2250.0,"mfuNominalPct":24.00088280547765,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.00088280547765,"vramAllocatedGb":168.019167744,"vramAllocatedPct":58.455958771213,"vramReservedGb":170.653646848,"vramReservedPct":59.37252682683918,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.398185984,"vramAllocatedPct":98.59775463807965,"vramReservedGb":283.753054208,"vramReservedPct":98.72121770809653,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.76 GiB is free. Including non-PyTorch memory, this process has 264.91 GiB memory in use. Of the allocated memory 263.94 GiB is allocated by PyTorch, and 158.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11794.70762633203,"meanTps":11761.599197260333,"stepMs":173.63719940185547,"jitter":0.0009529847192151667,"achievedTflops":162.9386207002926,"nominalPeakTflops":2250.0,"mfuNominalPct":7.24171647556856,"configuredPeakTflops":2250.0,"mfuConfiguredPct":7.24171647556856,"vramAllocatedGb":36.8966144,"vramAllocatedPct":12.836791177599228,"vramReservedGb":36.987469824,"vramReservedPct":12.868400909933916,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":18937.114649930463,"meanTps":18927.494341751044,"stepMs":216.29483032226562,"jitter":0.0009924199536398266,"achievedTflops":261.6077853608096,"nominalPeakTflops":2250.0,"mfuNominalPct":11.627012682702649,"configuredPeakTflops":2250.0,"mfuConfiguredPct":11.627012682702649,"vramAllocatedGb":45.638133248,"vramAllocatedPct":15.878074337357749,"vramReservedGb":46.080720896,"vramReservedPct":16.032056086294606,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":26743.57011185087,"meanTps":26741.173435384255,"stepMs":306.3166198730469,"jitter":0.0006538286822371927,"achievedTflops":369.4504827655218,"nominalPeakTflops":2250.0,"mfuNominalPct":16.420021456245415,"configuredPeakTflops":2250.0,"mfuConfiguredPct":16.420021456245415,"vramAllocatedGb":63.121138176,"vramAllocatedPct":21.960629256479926,"vramReservedGb":63.810043904,"vramReservedPct":22.200308129872386,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":33598.523762715355,"meanTps":33594.016985007765,"stepMs":487.6404724121094,"jitter":0.0003879309991993868,"achievedTflops":464.148607400905,"nominalPeakTflops":2250.0,"mfuNominalPct":20.62882699559578,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.62882699559578,"vramAllocatedGb":98.087148032,"vramAllocatedPct":34.12573909472429,"vramReservedGb":99.413393408,"vramReservedPct":34.58715636074771,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":39044.170483691414,"meanTps":38998.84518591448,"stepMs":839.2546081542969,"jitter":0.000698853021714542,"achievedTflops":539.377785914493,"nominalPeakTflops":2250.0,"mfuNominalPct":23.972346040644133,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.972346040644133,"vramAllocatedGb":168.019167744,"vramAllocatedPct":58.455958771213,"vramReservedGb":170.653646848,"vramReservedPct":59.37252682683918,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.398185984,"vramAllocatedPct":98.59775463807965,"vramReservedGb":283.753054208,"vramReservedPct":98.72121770809653,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.76 GiB is free. Including non-PyTorch memory, this process has 264.91 GiB memory in use. Of the allocated memory 263.94 GiB is allocated by PyTorch, and 158.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.525663232,"vramAllocatedPct":97.08885956913035,"vramReservedGb":24.717033472,"vramReservedPct":97.84643004464877,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 68.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 43.69 MiB is free. Including non-PyTorch memory, this process has 23.47 GiB memory in use. Of the allocated memory 22.84 GiB is allocated by PyTorch, and 178.50 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.525663232,"vramAllocatedPct":97.08885956913035,"vramReservedGb":24.717033472,"vramReservedPct":97.84643004464877,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 68.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 43.69 MiB is free. Including non-PyTorch memory, this process has 23.47 GiB memory in use. Of the allocated memory 22.84 GiB is allocated by PyTorch, and 178.50 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.940689408,"vramAllocatedPct":97.8372641830788,"vramReservedGb":33.021755392,"vramReservedPct":98.0780385638039,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 34.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 9.88 MiB is free. Including non-PyTorch memory, this process has 31.34 GiB memory in use. Of the allocated memory 30.68 GiB is allocated by PyTorch, and 73.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.940689408,"vramAllocatedPct":97.8372641830788,"vramReservedGb":33.021755392,"vramReservedPct":98.0780385638039,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 34.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 9.88 MiB is free. Including non-PyTorch memory, this process has 31.34 GiB memory in use. Of the allocated memory 30.68 GiB is allocated by PyTorch, and 73.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15666.079713705776,"meanTps":15657.610899542873,"stepMs":130.72830200195312,"jitter":0.0007192870307362434,"achievedTflops":216.41989790686108,"nominalPeakTflops":989.5,"mfuNominalPct":21.871642032022343,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.871642032022343,"vramAllocatedGb":36.946683904,"vramAllocatedPct":43.45774308467667,"vramReservedGb":37.050384384,"vramReservedPct":43.57971854610935,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":19720.631123773226,"meanTps":19718.93547533198,"stepMs":207.70126342773438,"jitter":0.0007174515699854024,"achievedTflops":272.43171568519347,"nominalPeakTflops":989.5,"mfuNominalPct":27.532260301687064,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.532260301687064,"vramAllocatedGb":45.688202752,"vramAllocatedPct":53.73976680440527,"vramReservedGb":46.101692416,"vramReservedPct":54.22612513721185,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":22721.133360557284,"meanTps":22708.96278351761,"stepMs":360.54539489746094,"jitter":0.0011963726574412484,"achievedTflops":313.8823146621668,"nominalPeakTflops":989.5,"mfuNominalPct":31.721305170507,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.721305170507,"vramAllocatedGb":63.17120768,"vramAllocatedPct":74.30377570120653,"vramReservedGb":63.831015424,"vramReservedPct":75.07986038308317,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.983931392,"vramAllocatedPct":98.78429477346168,"vramReservedGb":84.083212288,"vramReservedPct":98.90107179417605,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 138.19 MiB is free. Including non-PyTorch memory, this process has 79.04 GiB memory in use. Of the allocated memory 78.22 GiB is allocated by PyTorch, and 94.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":15631.11345326862,"meanTps":15631.347470873705,"stepMs":131.02073669433594,"jitter":0.0006704577552749495,"achievedTflops":215.93685462786044,"nominalPeakTflops":989.5,"mfuNominalPct":21.82282512661551,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.82282512661551,"vramAllocatedGb":36.946683904,"vramAllocatedPct":43.45774308467667,"vramReservedGb":37.050384384,"vramReservedPct":43.57971854610935,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":19704.077533671887,"meanTps":19701.28373312972,"stepMs":207.8757553100586,"jitter":0.00025454844406610697,"achievedTflops":272.2030352274661,"nominalPeakTflops":989.5,"mfuNominalPct":27.50914959347813,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.50914959347813,"vramAllocatedGb":45.688202752,"vramAllocatedPct":53.73976680440527,"vramReservedGb":46.101692416,"vramReservedPct":54.22612513721185,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":22695.50898376991,"meanTps":22697.636415133606,"stepMs":360.9524688720703,"jitter":0.0008036183057038903,"achievedTflops":313.5283253355709,"nominalPeakTflops":989.5,"mfuNominalPct":31.68553060490863,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.68553060490863,"vramAllocatedGb":63.17120768,"vramAllocatedPct":74.30377570120653,"vramReservedGb":63.831015424,"vramReservedPct":75.07986038308317,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.983931392,"vramAllocatedPct":98.78429477346168,"vramReservedGb":84.083212288,"vramReservedPct":98.90107179417605,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 138.19 MiB is free. Including non-PyTorch memory, this process has 79.04 GiB memory in use. Of the allocated memory 78.22 GiB is allocated by PyTorch, and 94.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":17014.52772049356,"meanTps":17012.251255900293,"stepMs":120.36772537231445,"jitter":0.0009753478429788072,"achievedTflops":235.04810517344407,"nominalPeakTflops":989.5,"mfuNominalPct":23.754229931626483,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.754229931626483,"vramAllocatedGb":36.946683904,"vramAllocatedPct":24.611308973697984,"vramReservedGb":37.050384384,"vramReservedPct":24.68038701492713,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":21413.24171846168,"meanTps":21408.80837519679,"stepMs":191.28350830078125,"jitter":0.0020644828045025044,"achievedTflops":295.814375469445,"nominalPeakTflops":989.5,"mfuNominalPct":29.895338602268318,"configuredPeakTflops":989.5,"mfuConfiguredPct":29.895338602268318,"vramAllocatedGb":45.688202752,"vramAllocatedPct":30.43430033678052,"vramReservedGb":46.101692416,"vramReservedPct":30.709738368095497,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":24173.121526473275,"meanTps":24162.339247829954,"stepMs":338.8887939453125,"jitter":0.0020975903733803716,"achievedTflops":333.94088300677885,"nominalPeakTflops":989.5,"mfuNominalPct":33.7484469941161,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.7484469941161,"vramAllocatedGb":63.17120768,"vramAllocatedPct":42.080261235184956,"vramReservedGb":63.831015424,"vramReservedPct":42.51977922529785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25911.649952047093,"meanTps":25903.655173860938,"stepMs":632.3024597167969,"jitter":0.003844588513108583,"achievedTflops":357.957877126992,"nominalPeakTflops":989.5,"mfuNominalPct":36.17563184709368,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.17563184709368,"vramAllocatedGb":98.137217536,"vramAllocatedPct":65.37218303199383,"vramReservedGb":99.455336448,"vramReservedPct":66.25022209746444,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.322983936,"vramAllocatedPct":98.80244719755564,"vramReservedGb":148.486750208,"vramReservedPct":98.91153688826064,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 827.44 MiB is free. Including non-PyTorch memory, this process has 138.99 GiB memory in use. Of the allocated memory 138.14 GiB is allocated by PyTorch, and 136.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":17084.726664726964,"meanTps":17080.92852045867,"stepMs":119.87314987182617,"jitter":0.0008333484374870222,"achievedTflops":236.01787225121984,"nominalPeakTflops":989.5,"mfuNominalPct":23.852235699971686,"configuredPeakTflops":989.5,"mfuConfiguredPct":23.852235699971686,"vramAllocatedGb":36.946683904,"vramAllocatedPct":24.611308973697984,"vramReservedGb":37.050384384,"vramReservedPct":24.68038701492713,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":21489.14911287514,"meanTps":21484.29902466242,"stepMs":190.60782623291016,"jitter":0.0016838131829323602,"achievedTflops":296.86300223820604,"nominalPeakTflops":989.5,"mfuNominalPct":30.001314021041537,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.001314021041537,"vramAllocatedGb":45.688202752,"vramAllocatedPct":30.43430033678052,"vramReservedGb":46.101692416,"vramReservedPct":30.709738368095497,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":24260.186316938347,"meanTps":24279.36990278306,"stepMs":337.67259216308594,"jitter":0.0028583802562988734,"achievedTflops":335.1436442213314,"nominalPeakTflops":989.5,"mfuNominalPct":33.86999941600116,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.86999941600116,"vramAllocatedGb":63.17120768,"vramAllocatedPct":42.080261235184956,"vramReservedGb":63.831015424,"vramReservedPct":42.51977922529785,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":26002.81101306076,"meanTps":26013.991789260264,"stepMs":630.0857238769531,"jitter":0.0030504150092346702,"achievedTflops":359.2172264905982,"nominalPeakTflops":989.5,"mfuNominalPct":36.30290313194524,"configuredPeakTflops":989.5,"mfuConfiguredPct":36.30290313194524,"vramAllocatedGb":98.137217536,"vramAllocatedPct":65.37218303199383,"vramReservedGb":99.455336448,"vramReservedPct":66.25022209746444,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.322983936,"vramAllocatedPct":98.80244719755564,"vramReservedGb":148.486750208,"vramReservedPct":98.91153688826064,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 827.44 MiB is free. Including non-PyTorch memory, this process has 138.99 GiB memory in use. Of the allocated memory 138.14 GiB is allocated by PyTorch, and 136.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":5127.315907447898,"meanTps":5128.708184319523,"stepMs":399.42926025390625,"jitter":0.002722225514340912,"achievedTflops":70.83158042757162,"nominalPeakTflops":364.2,"mfuNominalPct":19.448539381540808,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.690367520373503,"vramAllocatedGb":36.8966144,"vramAllocatedPct":72.52143207238261,"vramReservedGb":36.987469824,"vramReservedPct":72.70001120673331,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":5699.563297860007,"meanTps":5698.833010653708,"stepMs":718.6515502929688,"jitter":0.001459382471461147,"achievedTflops":78.73692267487982,"nominalPeakTflops":364.2,"mfuNominalPct":21.619144062295394,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.44153168310788,"vramAllocatedGb":45.638133248,"vramAllocatedPct":89.70315661957262,"vramReservedGb":46.080720896,"vramReservedPct":90.57307627405744,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":50.219517952,"vramAllocatedPct":98.70800936857142,"vramReservedGb":50.28970496,"vramReservedPct":98.84596409465696,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 68.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 34.50 MiB is free. Process 2308332 has 47.34 GiB memory in use. Of the allocated memory 46.77 GiB is allocated by PyTorch, and 66.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":5133.448317076693,"meanTps":5131.670604149243,"stepMs":398.9521026611328,"jitter":0.0017986906846147888,"achievedTflops":70.91629692908576,"nominalPeakTflops":364.2,"mfuNominalPct":19.47180036493294,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":15.709133627747828,"vramAllocatedGb":36.8966144,"vramAllocatedPct":72.52143207238261,"vramReservedGb":36.987469824,"vramReservedPct":72.70001120673331,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":5689.226727237483,"meanTps":5679.662316820278,"stepMs":719.9572448730469,"jitter":0.008075582567476782,"achievedTflops":78.5941275659748,"nominalPeakTflops":364.2,"mfuNominalPct":21.579936179564747,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":17.40990020283726,"vramAllocatedGb":45.638133248,"vramAllocatedPct":89.70315661957262,"vramReservedGb":46.080720896,"vramReservedPct":90.57307627405744,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":50.219517952,"vramAllocatedPct":98.70800936857142,"vramReservedGb":50.28970496,"vramReservedPct":98.84596409465696,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 68.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 34.50 MiB is free. Process 2312978 has 47.34 GiB memory in use. Of the allocated memory 46.77 GiB is allocated by PyTorch, and 66.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":3641.570300345372,"meanTps":3642.0994048054317,"stepMs":562.3947448730469,"jitter":0.0017081409818927161,"achievedTflops":50.30666810228924,"nominalPeakTflops":154.8,"mfuNominalPct":32.497847611297956,"configuredPeakTflops":180.6,"mfuConfiguredPct":27.855297952541108,"vramAllocatedGb":36.8966144,"vramAllocatedPct":72.29820065980408,"vramReservedGb":36.987469824,"vramReservedPct":72.47623010185998,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":4419.16521207973,"meanTps":4418.955567406874,"stepMs":926.8718872070312,"jitter":0.0005533193721083726,"achievedTflops":61.048794689530794,"nominalPeakTflops":154.8,"mfuNominalPct":39.437205871789914,"configuredPeakTflops":180.6,"mfuConfiguredPct":33.803319318677076,"vramAllocatedGb":45.638133248,"vramAllocatedPct":89.42703738429671,"vramReservedGb":46.080720896,"vramReservedPct":90.29427930079773,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.558732288,"vramAllocatedPct":99.06885581521377,"vramReservedGb":50.62524928,"vramReservedPct":99.1991945715768,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 108.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 52.25 MiB is free. Process 1348507 has 47.47 GiB memory in use. Of the allocated memory 47.09 GiB is allocated by PyTorch, and 63.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":3642.9182663171955,"meanTps":3642.6619197959876,"stepMs":562.1866455078125,"jitter":0.0016224976022884273,"achievedTflops":50.32528965045797,"nominalPeakTflops":154.8,"mfuNominalPct":32.509877035179564,"configuredPeakTflops":180.6,"mfuConfiguredPct":27.865608887296773,"vramAllocatedGb":36.8966144,"vramAllocatedPct":72.29820065980408,"vramReservedGb":36.987469824,"vramReservedPct":72.47623010185998,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":4412.957638802148,"meanTps":4413.239690662633,"stepMs":928.1756896972656,"jitter":0.0009238933348185916,"achievedTflops":60.963039835761705,"nominalPeakTflops":154.8,"mfuNominalPct":39.38180867943262,"configuredPeakTflops":180.6,"mfuConfiguredPct":33.75583601094225,"vramAllocatedGb":45.638133248,"vramAllocatedPct":89.42703738429671,"vramReservedGb":46.080720896,"vramReservedPct":90.29427930079773,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.558732288,"vramAllocatedPct":99.06885581521377,"vramReservedGb":50.62524928,"vramReservedPct":99.1991945715768,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 108.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 52.25 MiB is free. Process 1380474 has 47.47 GiB memory in use. Of the allocated memory 47.09 GiB is allocated by PyTorch, and 63.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":10293.852432880312,"meanTps":10294.089719219732,"stepMs":198.95369720458984,"jitter":0.00023440959803329022,"achievedTflops":142.20497618451546,"nominalPeakTflops":468.0,"mfuNominalPct":30.385678671905012,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.385678671905012,"vramAllocatedGb":36.8966144,"vramAllocatedPct":36.18239190706421,"vramReservedGb":36.987469824,"vramReservedPct":36.27148860635515,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":13036.221339398746,"meanTps":13034.17723128714,"stepMs":314.20147705078125,"jitter":0.0003893069997593182,"achievedTflops":180.08957843458774,"nominalPeakTflops":468.0,"mfuNominalPct":38.48067915268969,"configuredPeakTflops":468.0,"mfuConfiguredPct":38.48067915268969,"vramAllocatedGb":45.638133248,"vramAllocatedPct":44.75469768537769,"vramReservedGb":46.080720896,"vramReservedPct":45.18871798760796,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14083.26665126316,"meanTps":14081.615179159588,"stepMs":581.6832275390625,"jitter":0.0003859617611117721,"achievedTflops":194.55404201696743,"nominalPeakTflops":468.0,"mfuNominalPct":41.57137649935201,"configuredPeakTflops":468.0,"mfuConfiguredPct":41.57137649935201,"vramAllocatedGb":63.121138176,"vramAllocatedPct":61.89927710830791,"vramReservedGb":63.810043904,"vramReservedPct":62.57484741314101,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":14649.899513544591,"meanTps":14649.772870647794,"stepMs":1118.3694458007812,"jitter":0.00031920212253589884,"achievedTflops":202.3818220644757,"nominalPeakTflops":468.0,"mfuNominalPct":43.24397907360592,"configuredPeakTflops":468.0,"mfuConfiguredPct":43.24397907360592,"vramAllocatedGb":98.087148032,"vramAllocatedPct":96.18843595416836,"vramReservedGb":99.413393408,"vramReservedPct":97.4890086690287,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.99510272,"vramAllocatedPct":99.04010020250655,"vramReservedGb":101.133058048,"vramReservedPct":99.17538507415493,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 311.75 MiB is free. Including non-PyTorch memory, this process has 94.66 GiB memory in use. Of the allocated memory 93.90 GiB is allocated by PyTorch, and 111.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":10104.794637577335,"meanTps":10106.433684854355,"stepMs":202.67606353759766,"jitter":0.007669068035246318,"achievedTflops":139.59322713780466,"nominalPeakTflops":468.0,"mfuNominalPct":29.827612636283046,"configuredPeakTflops":468.0,"mfuConfiguredPct":29.827612636283046,"vramAllocatedGb":36.8966144,"vramAllocatedPct":36.18239190706421,"vramReservedGb":36.966498304,"vramReservedPct":36.25092304043897,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":13052.986130213763,"meanTps":13050.877434521617,"stepMs":313.7979278564453,"jitter":0.00017482588551524355,"achievedTflops":180.32117653589458,"nominalPeakTflops":468.0,"mfuNominalPct":38.5301659264732,"configuredPeakTflops":468.0,"mfuConfiguredPct":38.5301659264732,"vramAllocatedGb":45.638133248,"vramAllocatedPct":44.75469768537769,"vramReservedGb":46.059749376,"vramReservedPct":45.16815242169179,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14096.797970833355,"meanTps":14096.17948682873,"stepMs":581.1248779296875,"jitter":0.0003294960797802786,"achievedTflops":194.74097115644864,"nominalPeakTflops":468.0,"mfuNominalPct":41.611318623172785,"configuredPeakTflops":468.0,"mfuConfiguredPct":41.611318623172785,"vramAllocatedGb":63.121138176,"vramAllocatedPct":61.89927710830791,"vramReservedGb":63.789072384,"vramReservedPct":62.554281847224836,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":14643.531706068512,"meanTps":14643.04707482755,"stepMs":1118.8557739257812,"jitter":0.00021521613294570556,"achievedTflops":202.2938536467829,"nominalPeakTflops":468.0,"mfuNominalPct":43.22518240315874,"configuredPeakTflops":468.0,"mfuConfiguredPct":43.22518240315874,"vramAllocatedGb":98.087148032,"vramAllocatedPct":96.18843595416836,"vramReservedGb":99.413393408,"vramReservedPct":97.4890086690287,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.99510272,"vramAllocatedPct":99.04010020250655,"vramReservedGb":101.133058048,"vramReservedPct":99.17538507415493,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 293.75 MiB is free. Including non-PyTorch memory, this process has 94.68 GiB memory in use. Of the allocated memory 93.90 GiB is allocated by PyTorch, and 131.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":8403.811282467292,"meanTps":8402.260360232196,"stepMs":487.3979034423828,"jitter":0.0018892850938552615,"achievedTflops":130.89911990239415,"nominalPeakTflops":312.0,"mfuNominalPct":41.95484612256222,"configuredPeakTflops":312.0,"mfuConfiguredPct":41.95484612256222,"vramAllocatedGb":45.653100544,"vramAllocatedPct":53.72581229504151,"vramReservedGb":45.820674048,"vramReservedPct":53.92301735919371,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":9211.9824786802,"meanTps":9184.348916132853,"stepMs":889.2765502929688,"jitter":0.0033354642620574038,"achievedTflops":143.48732479645696,"nominalPeakTflops":312.0,"mfuNominalPct":45.98952717835159,"configuredPeakTflops":312.0,"mfuConfiguredPct":45.98952717835159,"vramAllocatedGb":63.13645824,"vramAllocatedPct":74.30070387238509,"vramReservedGb":63.875055616,"vramReservedPct":75.16990538360209,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.294309888,"vramAllocatedPct":99.1998400244022,"vramReservedGb":84.418756608,"vramReservedPct":99.34629231438434,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 84.75 MiB is free. Process 764857 has 79.05 GiB memory in use. Of the allocated memory 78.43 GiB is allocated by PyTorch, and 118.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":8411.721043631356,"meanTps":8409.476194014285,"stepMs":486.93959045410156,"jitter":0.0012703723258592444,"achievedTflops":131.0223236179719,"nominalPeakTflops":312.0,"mfuNominalPct":41.99433449293971,"configuredPeakTflops":312.0,"mfuConfiguredPct":41.99433449293971,"vramAllocatedGb":45.653100544,"vramAllocatedPct":53.72581229504151,"vramReservedGb":45.820674048,"vramReservedPct":53.92301735919371,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":9212.352682312625,"meanTps":9187.898490855625,"stepMs":889.2408142089844,"jitter":0.003889691163723719,"achievedTflops":143.49309114794207,"nominalPeakTflops":312.0,"mfuNominalPct":45.991375367930154,"configuredPeakTflops":312.0,"mfuConfiguredPct":45.991375367930154,"vramAllocatedGb":63.13645824,"vramAllocatedPct":74.30070387238509,"vramReservedGb":63.875055616,"vramReservedPct":75.16990538360209,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.294309888,"vramAllocatedPct":99.1998400244022,"vramReservedGb":84.418756608,"vramReservedPct":99.34629231438434,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 84.75 MiB is free. Process 852916 has 79.05 GiB memory in use. Of the allocated memory 78.43 GiB is allocated by PyTorch, and 118.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":34181.557837116525,"meanTps":34101.834758105346,"stepMs":479.3227996826172,"jitter":0.0007651605539495097,"achievedTflops":532.417457672575,"nominalPeakTflops":2250.0,"mfuNominalPct":23.66299811878111,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.66299811878111,"vramAllocatedGb":98.103108096,"vramAllocatedPct":51.227972475055644,"vramReservedGb":99.400810496,"vramReservedPct":51.90561321568295,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":36577.61035996185,"meanTps":36582.43876593982,"stepMs":895.8485717773438,"jitter":0.0009641012215703362,"achievedTflops":569.7387582037646,"nominalPeakTflops":2250.0,"mfuNominalPct":25.321722586833985,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.321722586833985,"vramAllocatedGb":168.036407808,"vramAllocatedPct":87.74609327944864,"vramReservedGb":170.64525824,"vramReservedPct":89.10839586818265,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.211226624,"vramAllocatedPct":99.32545126302828,"vramReservedGb":190.339612672,"vramReservedPct":99.39249253277775,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 640.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 575.81 MiB is free. Including non-PyTorch memory, this process has 177.78 GiB memory in use. Of the allocated memory 176.84 GiB is allocated by PyTorch, and 122.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":34310.46963154413,"meanTps":34295.99358450666,"stepMs":477.5218811035156,"jitter":0.0002669117121225668,"achievedTflops":534.4254085734735,"nominalPeakTflops":2250.0,"mfuNominalPct":23.75224038104327,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.75224038104327,"vramAllocatedGb":98.103108096,"vramAllocatedPct":51.227972475055644,"vramReservedGb":99.400810496,"vramReservedPct":51.90561321568295,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":36608.38592252539,"meanTps":36612.17627450514,"stepMs":895.095458984375,"jitter":0.0009430250734321228,"achievedTflops":570.2181233297371,"nominalPeakTflops":2250.0,"mfuNominalPct":25.343027703543868,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.343027703543868,"vramAllocatedGb":168.036407808,"vramAllocatedPct":87.74609327944864,"vramReservedGb":170.64525824,"vramReservedPct":89.10839586818265,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.211226624,"vramAllocatedPct":99.32545126302828,"vramReservedGb":190.339612672,"vramReservedPct":99.39249253277775,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 640.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 575.81 MiB is free. Including non-PyTorch memory, this process has 177.78 GiB memory in use. Of the allocated memory 176.84 GiB is allocated by PyTorch, and 122.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":30546.497400277694,"meanTps":30545.409790609057,"stepMs":536.3626403808594,"jitter":0.0005056293548115273,"achievedTflops":475.7971700458261,"nominalPeakTflops":2250.0,"mfuNominalPct":21.146540890925603,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.146540890925603,"vramAllocatedGb":98.103108096,"vramAllocatedPct":34.13129179954777,"vramReservedGb":99.400810496,"vramReservedPct":34.582778609119906,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":35122.68283797625,"meanTps":35109.23618605589,"stepMs":932.9583435058594,"jitter":0.000599612674408141,"achievedTflops":547.0765724706062,"nominalPeakTflops":2250.0,"mfuNominalPct":24.314514332026942,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.314514332026942,"vramAllocatedGb":168.036407808,"vramAllocatedPct":58.46195680396086,"vramReservedGb":170.64525824,"vramReservedPct":59.36960832575397,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.412866048,"vramAllocatedPct":98.60286201497875,"vramReservedGb":283.667070976,"vramReservedPct":98.6913030719732,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.77 GiB is free. Including non-PyTorch memory, this process has 264.91 GiB memory in use. Of the allocated memory 263.95 GiB is allocated by PyTorch, and 142.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":30527.745333064715,"meanTps":30509.954782865585,"stepMs":536.6921081542969,"jitter":0.0009261651736601963,"achievedTflops":475.5050848225833,"nominalPeakTflops":2250.0,"mfuNominalPct":21.133559325448147,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.133559325448147,"vramAllocatedGb":98.103108096,"vramAllocatedPct":34.13129179954777,"vramReservedGb":99.400810496,"vramReservedPct":34.582778609119906,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":35115.84487387502,"meanTps":35105.82511280246,"stepMs":933.1400146484375,"jitter":0.0006701418105530629,"achievedTflops":546.9700632389387,"nominalPeakTflops":2250.0,"mfuNominalPct":24.309780588397274,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.309780588397274,"vramAllocatedGb":168.036407808,"vramAllocatedPct":58.46195680396086,"vramReservedGb":170.64525824,"vramReservedPct":59.36960832575397,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.412866048,"vramAllocatedPct":98.60286201497875,"vramReservedGb":283.667070976,"vramReservedPct":98.6913030719732,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.77 GiB is free. Including non-PyTorch memory, this process has 264.91 GiB memory in use. Of the allocated memory 263.95 GiB is allocated by PyTorch, and 142.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":18383.18984834091,"meanTps":18379.93369705321,"stepMs":222.812255859375,"jitter":0.0007508834476172944,"achievedTflops":286.33952991861673,"nominalPeakTflops":989.5,"mfuNominalPct":28.93779989071417,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.93779989071417,"vramAllocatedGb":45.703170048,"vramAllocatedPct":53.75737176472946,"vramReservedGb":45.862617088,"vramReservedPct":53.944917919559934,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":21079.24647721147,"meanTps":21080.617465538526,"stepMs":388.6286926269531,"jitter":0.0006175663910523979,"achievedTflops":328.33374278991766,"nominalPeakTflops":989.5,"mfuNominalPct":33.181783000496985,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.181783000496985,"vramAllocatedGb":63.186527744,"vramAllocatedPct":74.32179559731095,"vramReservedGb":63.896027136,"vramReservedPct":75.15632901244466,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.998611456,"vramAllocatedPct":98.8015618833175,"vramReservedGb":84.125155328,"vramReservedPct":98.95040639376411,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 98.19 MiB is free. Including non-PyTorch memory, this process has 79.07 GiB memory in use. Of the allocated memory 78.23 GiB is allocated by PyTorch, and 120.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":18376.396651008632,"meanTps":18375.345274297197,"stepMs":222.89462280273438,"jitter":0.0008990930967408658,"achievedTflops":286.2337179813623,"nominalPeakTflops":989.5,"mfuNominalPct":28.927106415498972,"configuredPeakTflops":989.5,"mfuConfiguredPct":28.927106415498972,"vramAllocatedGb":45.703170048,"vramAllocatedPct":53.75737176472946,"vramReservedGb":45.862617088,"vramReservedPct":53.944917919559934,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":21063.827286247462,"meanTps":21066.81987431472,"stepMs":388.9131774902344,"jitter":0.00036553383511037866,"achievedTflops":328.0935709846552,"nominalPeakTflops":989.5,"mfuNominalPct":33.15751096358314,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.15751096358314,"vramAllocatedGb":63.186527744,"vramAllocatedPct":74.32179559731095,"vramReservedGb":63.896027136,"vramReservedPct":75.15632901244466,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.998611456,"vramAllocatedPct":98.8015618833175,"vramReservedGb":84.125155328,"vramReservedPct":98.95040639376411,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 98.19 MiB is free. Including non-PyTorch memory, this process has 79.07 GiB memory in use. Of the allocated memory 78.23 GiB is allocated by PyTorch, and 120.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":22456.44261892364,"meanTps":22447.00832984248,"stepMs":364.7950897216797,"jitter":0.0027860751971444933,"achievedTflops":349.7851720074202,"nominalPeakTflops":989.5,"mfuNominalPct":35.34968893455485,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.34968893455485,"vramAllocatedGb":63.186527744,"vramAllocatedPct":42.090466395398536,"vramReservedGb":63.883444224,"vramReservedPct":42.554703642311104,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23930.2537720496,"meanTps":23933.859935609034,"stepMs":684.6563415527344,"jitter":0.00270788980289968,"achievedTflops":372.7414922247729,"nominalPeakTflops":989.5,"mfuNominalPct":37.669680871629396,"configuredPeakTflops":989.5,"mfuConfiguredPct":37.669680871629396,"vramAllocatedGb":98.1531776,"vramAllocatedPct":65.38281451565727,"vramReservedGb":99.514056704,"vramReservedPct":66.28933744451929,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.337664,"vramAllocatedPct":98.81222603431935,"vramReservedGb":148.524498944,"vramReservedPct":98.93668246851018,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 793.44 MiB is free. Including non-PyTorch memory, this process has 139.03 GiB memory in use. Of the allocated memory 138.15 GiB is allocated by PyTorch, and 158.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":22362.492357123985,"meanTps":22360.93308595334,"stepMs":366.3276824951172,"jitter":0.0030704124031126576,"achievedTflops":348.3217875773305,"nominalPeakTflops":989.5,"mfuNominalPct":35.20179763287827,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.20179763287827,"vramAllocatedGb":63.186527744,"vramAllocatedPct":42.090466395398536,"vramReservedGb":63.883444224,"vramReservedPct":42.554703642311104,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23880.49763266015,"meanTps":23901.663806206547,"stepMs":686.0828552246094,"jitter":0.0024117613441734455,"achievedTflops":371.96648257297267,"nominalPeakTflops":989.5,"mfuNominalPct":37.591357511164496,"configuredPeakTflops":989.5,"mfuConfiguredPct":37.591357511164496,"vramAllocatedGb":98.1531776,"vramAllocatedPct":65.38281451565727,"vramReservedGb":99.514056704,"vramReservedPct":66.28933744451929,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.337664,"vramAllocatedPct":98.81222603431935,"vramReservedGb":148.524498944,"vramReservedPct":98.93668246851018,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 793.44 MiB is free. Including non-PyTorch memory, this process has 139.03 GiB memory in use. Of the allocated memory 138.15 GiB is allocated by PyTorch, and 158.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":5509.341677528939,"meanTps":5507.223501167258,"stepMs":743.4645080566406,"jitter":0.002405391094147426,"achievedTflops":85.8143945158165,"nominalPeakTflops":364.2,"mfuNominalPct":23.562436714941377,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":19.009308847319854,"vramAllocatedGb":45.653100544,"vramAllocatedPct":89.73257530087503,"vramReservedGb":45.820674048,"vramReservedPct":90.06194618449375,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":50.150311936,"vramAllocatedPct":98.57198281247786,"vramReservedGb":50.291802112,"vramReservedPct":98.85008611150829,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 80.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 32.50 MiB is free. Process 2319972 has 47.34 GiB memory in use. Of the allocated memory 46.71 GiB is allocated by PyTorch, and 134.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":4228.28207838042,"meanTps":4229.027514092759,"stepMs":968.7149353027344,"jitter":0.0007941986811455593,"achievedTflops":65.86040359018703,"nominalPeakTflops":154.8,"mfuNominalPct":42.54548035541798,"configuredPeakTflops":180.6,"mfuConfiguredPct":36.46755459035827,"vramAllocatedGb":45.653100544,"vramAllocatedPct":89.45636551066114,"vramReservedGb":45.820674048,"vramReservedPct":89.78472254326353,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.346919936,"vramAllocatedPct":98.65381361754478,"vramReservedGb":50.501517312,"vramReservedPct":98.95674417887908,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 170.25 MiB is free. Process 1415479 has 47.36 GiB memory in use. Of the allocated memory 46.89 GiB is allocated by PyTorch, and 147.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":4227.892225542559,"meanTps":4227.554988249076,"stepMs":968.8042602539062,"jitter":0.0008913595880371871,"achievedTflops":65.85433117950903,"nominalPeakTflops":154.8,"mfuNominalPct":42.54155760950196,"configuredPeakTflops":180.6,"mfuConfiguredPct":36.46419223671597,"vramAllocatedGb":45.653100544,"vramAllocatedPct":89.45636551066114,"vramReservedGb":45.820674048,"vramReservedPct":89.78472254326353,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.346919936,"vramAllocatedPct":98.65381361754478,"vramReservedGb":50.501517312,"vramReservedPct":98.95674417887908,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 170.25 MiB is free. Process 1449495 has 47.36 GiB memory in use. Of the allocated memory 46.89 GiB is allocated by PyTorch, and 147.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":13491.454136898961,"meanTps":13489.822479293245,"stepMs":607.1991882324219,"jitter":0.00021467730608432965,"achievedTflops":210.14506553806143,"nominalPeakTflops":468.0,"mfuNominalPct":44.902791781637056,"configuredPeakTflops":468.0,"mfuConfiguredPct":44.902791781637056,"vramAllocatedGb":63.13645824,"vramAllocatedPct":61.91430061571376,"vramReservedGb":63.841501184,"vramReservedPct":62.60569576201527,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":14040.009893656788,"meanTps":14039.112298450682,"stepMs":1166.9507446289062,"jitter":0.00024108614526635125,"achievedTflops":218.68945847639378,"nominalPeakTflops":468.0,"mfuNominalPct":46.72851676846021,"configuredPeakTflops":468.0,"mfuConfiguredPct":46.72851676846021,"vramAllocatedGb":98.103108096,"vramAllocatedPct":96.20408707283873,"vramReservedGb":99.451142144,"vramReservedPct":97.52602668767781,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.842010624,"vramAllocatedPct":98.8899715713185,"vramReservedGb":101.149835264,"vramReservedPct":99.19183752688787,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 135.75 MiB is free. Including non-PyTorch memory, this process has 94.83 GiB memory in use. Of the allocated memory 93.92 GiB is allocated by PyTorch, and 273.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":13478.155293129987,"meanTps":13478.35845315956,"stepMs":607.7983093261719,"jitter":0.000874835415250848,"achievedTflops":209.93792060267833,"nominalPeakTflops":468.0,"mfuNominalPct":44.85853004330734,"configuredPeakTflops":468.0,"mfuConfiguredPct":44.85853004330734,"vramAllocatedGb":63.30224896,"vramAllocatedPct":62.07688205856816,"vramReservedGb":69.319262208,"vramReservedPct":67.97742157931978,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":98.272946688,"vramAllocatedPct":96.37063803141905,"vramReservedGb":100.828971008,"vramReservedPct":98.87718436837042,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 94.97 GiB of which 1.39 GiB is free. Including non-PyTorch memory, this process has 93.57 GiB memory in use. Of the allocated memory 89.57 GiB is allocated by PyTorch, and 3.36 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":8262.728939549868,"meanTps":8263.153193092941,"stepMs":991.4400024414062,"jitter":0.0006425093203522609,"achievedTflops":157.81297209069072,"nominalPeakTflops":312.0,"mfuNominalPct":50.58108079829831,"configuredPeakTflops":312.0,"mfuConfiguredPct":50.58108079829831,"vramAllocatedGb":63.166072832,"vramAllocatedPct":74.33555513094205,"vramReservedGb":63.346573312,"vramReservedPct":74.54797301257747,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.239783936,"vramAllocatedPct":99.1356723988203,"vramReservedGb":84.374716416,"vramReservedPct":99.29446461679896,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 80.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 46.75 MiB is free. Process 764857 has 79.08 GiB memory in use. Of the allocated memory 78.45 GiB is allocated by PyTorch, and 128.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":8265.931502400359,"meanTps":8267.125942790606,"stepMs":991.0558776855469,"jitter":0.0007131106932091193,"achievedTflops":157.87413904478552,"nominalPeakTflops":312.0,"mfuNominalPct":50.60068559127741,"configuredPeakTflops":312.0,"mfuConfiguredPct":50.60068559127741,"vramAllocatedGb":63.166072832,"vramAllocatedPct":74.33555513094205,"vramReservedGb":63.346573312,"vramReservedPct":74.54797301257747,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.239783936,"vramAllocatedPct":99.1356723988203,"vramReservedGb":84.374716416,"vramReservedPct":99.29446461679896,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 80.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 46.75 MiB is free. Process 852916 has 79.08 GiB memory in use. Of the allocated memory 78.45 GiB is allocated by PyTorch, and 128.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":28324.261516580744,"meanTps":28296.080615005125,"stepMs":578.4440307617188,"jitter":0.0011085990553987995,"achievedTflops":540.9757387550333,"nominalPeakTflops":2250.0,"mfuNominalPct":24.043366166890365,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.043366166890365,"vramAllocatedGb":98.133108224,"vramAllocatedPct":51.24363809219316,"vramReservedGb":99.428073472,"vramReservedPct":51.91984953307617,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30064.959398905678,"meanTps":30054.680032850887,"stepMs":1089.9066772460938,"jitter":0.0008744310966363893,"achievedTflops":574.2219832260075,"nominalPeakTflops":2250.0,"mfuNominalPct":25.520977032267,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.520977032267,"vramAllocatedGb":168.067047936,"vramAllocatedPct":87.76209309499251,"vramReservedGb":170.693492736,"vramReservedPct":89.13358319895528,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.240586752,"vramAllocatedPct":99.34078268175945,"vramReservedGb":190.3689728,"vramReservedPct":99.40782395150892,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 640.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 547.81 MiB is free. Including non-PyTorch memory, this process has 177.80 GiB memory in use. Of the allocated memory 176.86 GiB is allocated by PyTorch, and 122.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":28258.167850400132,"meanTps":28188.34531223526,"stepMs":579.7969665527344,"jitter":0.000624214451164698,"achievedTflops":539.7133909311314,"nominalPeakTflops":2250.0,"mfuNominalPct":23.987261819161393,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.987261819161393,"vramAllocatedGb":98.133108224,"vramAllocatedPct":51.24363809219316,"vramReservedGb":99.428073472,"vramReservedPct":51.91984953307617,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":30075.622356546242,"meanTps":30075.747035426095,"stepMs":1089.520263671875,"jitter":0.0010194240225873494,"achievedTflops":574.425639070081,"nominalPeakTflops":2250.0,"mfuNominalPct":25.530028403114713,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.530028403114713,"vramAllocatedGb":168.067047936,"vramAllocatedPct":87.76209309499251,"vramReservedGb":170.693492736,"vramReservedPct":89.13358319895528,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.240586752,"vramAllocatedPct":99.34078268175945,"vramReservedGb":190.3689728,"vramReservedPct":99.40782395150892,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 640.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 547.81 MiB is free. Including non-PyTorch memory, this process has 177.80 GiB memory in use. Of the allocated memory 176.86 GiB is allocated by PyTorch, and 122.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25835.621779305064,"meanTps":25792.922736118893,"stepMs":634.1631774902344,"jitter":0.0002035517362897161,"achievedTflops":493.44427109153435,"nominalPeakTflops":2250.0,"mfuNominalPct":21.93085649295708,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.93085649295708,"vramAllocatedGb":98.133108224,"vramAllocatedPct":34.14172921730817,"vramReservedGb":99.428073472,"vramReservedPct":34.59226373764682,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":29187.084616034488,"meanTps":29186.36916296781,"stepMs":1122.6883544921875,"jitter":0.0001214669400437944,"achievedTflops":557.4551221051925,"nominalPeakTflops":2250.0,"mfuNominalPct":24.775783204675218,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.775783204675218,"vramAllocatedGb":168.067047936,"vramAllocatedPct":58.47261688568345,"vramReservedGb":170.693492736,"vramReservedPct":59.38638970699389,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.442226176,"vramAllocatedPct":98.61307676877696,"vramReservedGb":283.696431104,"vramReservedPct":98.70151782577142,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.74 GiB is free. Including non-PyTorch memory, this process has 264.94 GiB memory in use. Of the allocated memory 263.98 GiB is allocated by PyTorch, and 142.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25809.54924297582,"meanTps":25807.850121574826,"stepMs":634.8038024902344,"jitter":0.0017314728715114298,"achievedTflops":492.94630190022207,"nominalPeakTflops":2250.0,"mfuNominalPct":21.90872452889876,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.90872452889876,"vramAllocatedGb":98.133108224,"vramAllocatedPct":34.14172921730817,"vramReservedGb":99.428073472,"vramReservedPct":34.59226373764682,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":29171.560072951866,"meanTps":29165.1842401058,"stepMs":1123.2858276367188,"jitter":0.0006178065590935407,"achievedTflops":557.1586130096937,"nominalPeakTflops":2250.0,"mfuNominalPct":24.76260502265305,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.76260502265305,"vramAllocatedGb":168.067047936,"vramAllocatedPct":58.47261688568345,"vramReservedGb":170.693492736,"vramReservedPct":59.38638970699389,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.442226176,"vramAllocatedPct":98.61307676877696,"vramReservedGb":283.696431104,"vramReservedPct":98.70151782577142,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.74 GiB is free. Including non-PyTorch memory, this process has 264.94 GiB memory in use. Of the allocated memory 263.98 GiB is allocated by PyTorch, and 142.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":18561.060346524628,"meanTps":18559.329571795126,"stepMs":441.3540954589844,"jitter":0.000661631023743057,"achievedTflops":354.5046823960428,"nominalPeakTflops":989.5,"mfuNominalPct":35.826648044066985,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.826648044066985,"vramAllocatedGb":63.216142336,"vramAllocatedPct":74.35662912483504,"vramReservedGb":63.409487872,"vramReservedPct":74.5840476572232,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.027971584,"vramAllocatedPct":98.83609610302915,"vramReservedGb":84.185972736,"vramReservedPct":99.02194156316679,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 40.19 MiB is free. Including non-PyTorch memory, this process has 79.13 GiB memory in use. Of the allocated memory 78.26 GiB is allocated by PyTorch, and 150.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":18562.64806147602,"meanTps":18555.771570479246,"stepMs":441.31634521484375,"jitter":0.0010423579221818823,"achievedTflops":354.53500676189634,"nominalPeakTflops":989.5,"mfuNominalPct":35.82971265911029,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.82971265911029,"vramAllocatedGb":63.216142336,"vramAllocatedPct":74.35662912483504,"vramReservedGb":63.409487872,"vramReservedPct":74.5840476572232,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.027971584,"vramAllocatedPct":98.83609610302915,"vramReservedGb":84.185972736,"vramReservedPct":99.02194156316679,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 40.19 MiB is free. Including non-PyTorch memory, this process has 79.13 GiB memory in use. Of the allocated memory 78.26 GiB is allocated by PyTorch, and 150.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":19564.629766239752,"meanTps":19575.12779962191,"stepMs":418.7147979736328,"jitter":0.0038636033455616767,"achievedTflops":373.67223272756814,"nominalPeakTflops":989.5,"mfuNominalPct":37.76374256973907,"configuredPeakTflops":989.5,"mfuConfiguredPct":37.76374256973907,"vramAllocatedGb":63.216142336,"vramAllocatedPct":42.11019357512962,"vramReservedGb":63.388516352,"vramReservedPct":42.225017145705976,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":20675.10521614685,"meanTps":20671.567321832605,"stepMs":792.4506225585938,"jitter":0.006701012884727405,"achievedTflops":394.88162159481726,"nominalPeakTflops":989.5,"mfuNominalPct":39.90718762959245,"configuredPeakTflops":989.5,"mfuConfiguredPct":39.90718762959245,"vramAllocatedGb":98.183177728,"vramAllocatedPct":65.40279851263456,"vramReservedGb":99.558096896,"vramReservedPct":66.31867395481042,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.367024128,"vramAllocatedPct":98.83178370784677,"vramReservedGb":148.568539136,"vramReservedPct":98.96601897880132,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 751.44 MiB is free. Including non-PyTorch memory, this process has 139.07 GiB memory in use. Of the allocated memory 138.18 GiB is allocated by PyTorch, and 172.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":19564.657572246113,"meanTps":19559.647778146995,"stepMs":418.7142028808594,"jitter":0.002654063164360701,"achievedTflops":373.6727638049565,"nominalPeakTflops":989.5,"mfuNominalPct":37.76379624102643,"configuredPeakTflops":989.5,"mfuConfiguredPct":37.76379624102643,"vramAllocatedGb":63.216142336,"vramAllocatedPct":42.11019357512962,"vramReservedGb":63.388516352,"vramReservedPct":42.225017145705976,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":20678.067527774394,"meanTps":20692.670822001164,"stepMs":792.3370971679688,"jitter":0.005037229517262481,"achievedTflops":394.9381999003168,"nominalPeakTflops":989.5,"mfuNominalPct":39.912905497758146,"configuredPeakTflops":989.5,"mfuConfiguredPct":39.912905497758146,"vramAllocatedGb":98.183177728,"vramAllocatedPct":65.40279851263456,"vramReservedGb":99.558096896,"vramReservedPct":66.31867395481042,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.367024128,"vramAllocatedPct":98.83178370784677,"vramReservedGb":148.568539136,"vramReservedPct":98.96601897880132,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 751.44 MiB is free. Including non-PyTorch memory, this process has 139.07 GiB memory in use. Of the allocated memory 138.18 GiB is allocated by PyTorch, and 172.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":50.179672064,"vramAllocatedPct":98.62969104839634,"vramReservedGb":50.32116224,"vramReservedPct":98.90779434742677,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 80.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 4.50 MiB is free. Process 2319972 has 47.37 GiB memory in use. Of the allocated memory 46.73 GiB is allocated by PyTorch, and 134.94 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.376280064,"vramAllocatedPct":98.71134421920186,"vramReservedGb":50.61476352,"vramReservedPct":99.17864792812784,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 162.25 MiB is free. Process 1415479 has 47.36 GiB memory in use. Of the allocated memory 46.92 GiB is allocated by PyTorch, and 127.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.376280064,"vramAllocatedPct":98.71134421920186,"vramReservedGb":50.61476352,"vramReservedPct":99.17864792812784,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 162.25 MiB is free. Process 1449495 has 47.36 GiB memory in use. Of the allocated memory 46.92 GiB is allocated by PyTorch, and 127.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":12502.57475240114,"meanTps":12502.336181913692,"stepMs":655.2250366210938,"jitter":0.0001180809089195951,"achievedTflops":238.79138416586412,"nominalPeakTflops":468.0,"mfuNominalPct":51.02380003544105,"configuredPeakTflops":468.0,"mfuConfiguredPct":51.02380003544105,"vramAllocatedGb":63.166072832,"vramAllocatedPct":61.943341946235186,"vramReservedGb":63.346573312,"vramReservedPct":62.120348406393575,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":12977.393745790096,"meanTps":12977.916003227727,"stepMs":1262.5031127929688,"jitter":0.00011210070597076092,"achievedTflops":247.86013095642548,"nominalPeakTflops":468.0,"mfuNominalPct":52.96156644368066,"configuredPeakTflops":468.0,"mfuConfiguredPct":52.96156644368066,"vramAllocatedGb":98.133108224,"vramAllocatedPct":96.23350647638591,"vramReservedGb":99.495182336,"vramReservedPct":97.56921437610178,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.871370752,"vramAllocatedPct":98.91876336360113,"vramReservedGb":101.193875456,"vramReservedPct":99.23502521531184,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 93.75 MiB is free. Including non-PyTorch memory, this process has 94.87 GiB memory in use. Of the allocated memory 93.94 GiB is allocated by PyTorch, and 287.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":12486.260345912106,"meanTps":12484.345517245087,"stepMs":656.0811462402344,"jitter":0.00042384694685439656,"achievedTflops":238.479789171672,"nominalPeakTflops":468.0,"mfuNominalPct":50.95721990847692,"configuredPeakTflops":468.0,"mfuConfiguredPct":50.95721990847692,"vramAllocatedGb":63.335213568,"vramAllocatedPct":62.10920855749264,"vramReservedGb":68.82852864,"vramReservedPct":67.49618733688132,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":98.305714688,"vramAllocatedPct":96.40277172816306,"vramReservedGb":100.866719744,"vramReservedPct":98.91420238701953,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 94.97 GiB of which 385.75 MiB is free. Including non-PyTorch memory, this process has 94.59 GiB memory in use. Of the allocated memory 89.60 GiB is allocated by PyTorch, and 4.34 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.298504192,"vramAllocatedPct":99.20477599560081,"vramReservedGb":84.374716416,"vramReservedPct":99.29446461679896,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 80.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 46.75 MiB is free. Process 925444 has 79.08 GiB memory in use. Of the allocated memory 78.51 GiB is allocated by PyTorch, and 72.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.298504192,"vramAllocatedPct":99.20477599560081,"vramReservedGb":84.374716416,"vramReservedPct":99.29446461679896,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 80.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 46.75 MiB is free. Process 937161 has 79.08 GiB memory in use. Of the allocated memory 78.51 GiB is allocated by PyTorch, and 72.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21236.641215334912,"meanTps":21223.538912820284,"stepMs":771.4967651367188,"jitter":0.00042307375917934227,"achievedTflops":555.249130189583,"nominalPeakTflops":2250.0,"mfuNominalPct":24.67773911953702,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.67773911953702,"vramAllocatedGb":98.192017408,"vramAllocatedPct":51.27439958502504,"vramReservedGb":98.446606336,"vramReservedPct":51.40734210692022,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":22200.983587047107,"meanTps":22198.14969121317,"stepMs":1475.9706420898438,"jitter":0.0004519721384683667,"achievedTflops":580.4626400694553,"nominalPeakTflops":2250.0,"mfuNominalPct":25.79833955864246,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.79833955864246,"vramAllocatedGb":168.126408192,"vramAllocatedPct":87.79309013086119,"vramReservedGb":170.7606016,"vramReservedPct":89.16862644176936,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.299307008,"vramAllocatedPct":99.37144551922177,"vramReservedGb":190.540939264,"vramReservedPct":99.49762226122,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 640.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 383.81 MiB is free. Including non-PyTorch memory, this process has 177.96 GiB memory in use. Of the allocated memory 176.92 GiB is allocated by PyTorch, and 230.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21188.33710848736,"meanTps":21181.215355848864,"stepMs":773.2555847167969,"jitter":0.0004164942204090853,"achievedTflops":553.9861803172499,"nominalPeakTflops":2250.0,"mfuNominalPct":24.621608014099998,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.621608014099998,"vramAllocatedGb":98.192017408,"vramAllocatedPct":51.27439958502504,"vramReservedGb":98.446606336,"vramReservedPct":51.40734210692022,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":22204.60136196979,"meanTps":22189.966005973853,"stepMs":1475.7301635742188,"jitter":0.0005296760082558112,"achievedTflops":580.5572297156557,"nominalPeakTflops":2250.0,"mfuNominalPct":25.80254354291803,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.80254354291803,"vramAllocatedGb":168.126408192,"vramAllocatedPct":87.79309013086119,"vramReservedGb":170.7606016,"vramReservedPct":89.16862644176936,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.299307008,"vramAllocatedPct":99.37144551922177,"vramReservedGb":190.540939264,"vramReservedPct":99.49762226122,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 640.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 383.81 MiB is free. Including non-PyTorch memory, this process has 177.96 GiB memory in use. Of the allocated memory 176.92 GiB is allocated by PyTorch, and 230.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19932.16355996665,"meanTps":19932.8637103127,"stepMs":821.988037109375,"jitter":0.0004292867500103212,"achievedTflops":521.1425087068999,"nominalPeakTflops":2250.0,"mfuNominalPct":23.16188927586222,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.16188927586222,"vramAllocatedGb":98.192017408,"vramAllocatedPct":34.16222445530624,"vramReservedGb":98.446606336,"vramReservedPct":34.250799110677995,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":21879.187494419362,"meanTps":21877.385893542425,"stepMs":1497.6790161132812,"jitter":0.00019642749627281765,"achievedTflops":572.0490214224097,"nominalPeakTflops":2250.0,"mfuNominalPct":25.4244009521071,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.4244009521071,"vramAllocatedGb":168.126408192,"vramAllocatedPct":58.49326905724207,"vramReservedGb":170.7606016,"vramReservedPct":59.40973771567552,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.500946432,"vramAllocatedPct":98.63350627637338,"vramReservedGb":284.182970368,"vramReservedPct":98.87079088871322,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.60 GiB is free. Including non-PyTorch memory, this process has 265.08 GiB memory in use. Of the allocated memory 264.03 GiB is allocated by PyTorch, and 230.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":19919.894310868258,"meanTps":19916.82004852555,"stepMs":822.4943237304688,"jitter":0.00024330895348510177,"achievedTflops":520.8217192835215,"nominalPeakTflops":2250.0,"mfuNominalPct":23.14763196815651,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.14763196815651,"vramAllocatedGb":98.192017408,"vramAllocatedPct":34.16222445530624,"vramReservedGb":98.446606336,"vramReservedPct":34.250799110677995,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":21874.68026053345,"meanTps":21873.87635664393,"stepMs":1497.9876098632812,"jitter":0.00021432826404597412,"achievedTflops":571.9311761535023,"nominalPeakTflops":2250.0,"mfuNominalPct":25.419163384600104,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.419163384600104,"vramAllocatedGb":168.126408192,"vramAllocatedPct":58.49326905724207,"vramReservedGb":170.7606016,"vramReservedPct":59.40973771567552,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.500946432,"vramAllocatedPct":98.63350627637338,"vramReservedGb":284.182970368,"vramReservedPct":98.87079088871322,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.60 GiB is free. Including non-PyTorch memory, this process has 265.08 GiB memory in use. Of the allocated memory 264.03 GiB is allocated by PyTorch, and 230.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.08669184,"vramAllocatedPct":98.90516454245243,"vramReservedGb":84.165001216,"vramReservedPct":98.99727426337276,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 62.19 MiB is free. Including non-PyTorch memory, this process has 79.11 GiB memory in use. Of the allocated memory 78.31 GiB is allocated by PyTorch, and 74.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_16k_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":84.08669184,"vramAllocatedPct":98.90516454245243,"vramReservedGb":84.165001216,"vramReservedPct":98.99727426337276,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 216.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 62.19 MiB is free. Including non-PyTorch memory, this process has 79.11 GiB memory in use. Of the allocated memory 78.31 GiB is allocated by PyTorch, and 74.68 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":16589.14947806546,"meanTps":16586.033405920494,"stepMs":987.6335144042969,"jitter":0.0021265422819891065,"achievedTflops":433.73670651974413,"nominalPeakTflops":989.5,"mfuNominalPct":43.833926884259135,"configuredPeakTflops":989.5,"mfuConfiguredPct":43.833926884259135,"vramAllocatedGb":98.242086912,"vramAllocatedPct":65.44203971037182,"vramReservedGb":98.488549376,"vramReservedPct":65.60621584774002,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.425744384,"vramAllocatedPct":98.87089905490161,"vramReservedGb":148.65661952,"vramReservedPct":99.02469199938358,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 727.44 MiB is free. Including non-PyTorch memory, this process has 139.09 GiB memory in use. Of the allocated memory 138.23 GiB is allocated by PyTorch, and 140.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":16580.908407522686,"meanTps":16577.662696126554,"stepMs":988.1243896484375,"jitter":0.0019432909002972925,"achievedTflops":433.5212370768925,"nominalPeakTflops":989.5,"mfuNominalPct":43.81215129630041,"configuredPeakTflops":989.5,"mfuConfiguredPct":43.81215129630041,"vramAllocatedGb":98.242086912,"vramAllocatedPct":65.44203971037182,"vramReservedGb":98.488549376,"vramReservedPct":65.60621584774002,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.425744384,"vramAllocatedPct":98.87089905490161,"vramReservedGb":148.65661952,"vramReservedPct":99.02469199938358,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 727.44 MiB is free. Including non-PyTorch memory, this process has 139.09 GiB memory in use. Of the allocated memory 138.23 GiB is allocated by PyTorch, and 140.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":98.33599232,"vramAllocatedPct":96.43246326395455,"vramReservedGb":100.342431744,"vramReservedPct":98.4000632391152,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 94.97 GiB of which 907.75 MiB is free. Including non-PyTorch memory, this process has 94.08 GiB memory in use. Of the allocated memory 89.63 GiB is allocated by PyTorch, and 3.80 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":98.33599232,"vramAllocatedPct":96.43246326395455,"vramReservedGb":100.342431744,"vramReservedPct":98.4000632391152,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1.95 GiB. GPU 0 has a total capacity of 94.97 GiB of which 907.75 MiB is free. Including non-PyTorch memory, this process has 94.08 GiB memory in use. Of the allocated memory 89.63 GiB is allocated by PyTorch, and 3.80 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":14645.135922899506,"meanTps":14644.701759015357,"stepMs":2237.4664306640625,"jitter":0.0031042726777991,"achievedTflops":589.3007739325355,"nominalPeakTflops":2250.0,"mfuNominalPct":26.19114550811269,"configuredPeakTflops":2250.0,"mfuConfiguredPct":26.19114550811269,"vramAllocatedGb":168.24390656,"vramAllocatedPct":87.85444601732176,"vramReservedGb":168.78927872,"vramReservedPct":88.1392311841057,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.0812032,"vramAllocatedPct":99.257554980076,"vramReservedGb":190.352195584,"vramReservedPct":99.3990631408054,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 243.81 MiB is free. Including non-PyTorch memory, this process has 178.10 GiB memory in use. Of the allocated memory 177.03 GiB is allocated by PyTorch, and 258.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":14682.327716699134,"meanTps":14674.380807207486,"stepMs":2231.7987060546875,"jitter":0.0008770931299446832,"achievedTflops":590.7973222053166,"nominalPeakTflops":2250.0,"mfuNominalPct":26.257658764680738,"configuredPeakTflops":2250.0,"mfuConfiguredPct":26.257658764680738,"vramAllocatedGb":168.24390656,"vramAllocatedPct":87.85444601732176,"vramReservedGb":168.78927872,"vramReservedPct":88.1392311841057,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.0812032,"vramAllocatedPct":99.257554980076,"vramReservedGb":190.352195584,"vramReservedPct":99.3990631408054,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 320.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 243.75 MiB is free. Including non-PyTorch memory, this process has 178.10 GiB memory in use. Of the allocated memory 177.03 GiB is allocated by PyTorch, and 258.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":14682.15104483016,"meanTps":14680.241332727725,"stepMs":2231.8255615234375,"jitter":0.0002280284951442902,"achievedTflops":590.7902131644947,"nominalPeakTflops":2250.0,"mfuNominalPct":26.257342807310877,"configuredPeakTflops":2250.0,"mfuConfiguredPct":26.257342807310877,"vramAllocatedGb":168.24390656,"vramAllocatedPct":58.534148201257096,"vramReservedGb":168.78927872,"vramReservedPct":58.72388996065268,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.618386944,"vramAllocatedPct":98.67436529156623,"vramReservedGb":284.392685568,"vramReservedPct":98.94375341584332,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.50 GiB is free. Including non-PyTorch memory, this process has 265.17 GiB memory in use. Of the allocated memory 264.14 GiB is allocated by PyTorch, and 218.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":14687.48760701737,"meanTps":14687.300283681874,"stepMs":2231.0146484375,"jitter":0.0002658324976505312,"achievedTflops":591.0049493228765,"nominalPeakTflops":2250.0,"mfuNominalPct":26.26688663657229,"configuredPeakTflops":2250.0,"mfuConfiguredPct":26.26688663657229,"vramAllocatedGb":168.24390656,"vramAllocatedPct":58.534148201257096,"vramReservedGb":168.78927872,"vramReservedPct":58.72388996065268,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":283.618386944,"vramAllocatedPct":98.67436529156623,"vramReservedGb":284.392685568,"vramReservedPct":98.94375341584332,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 267.69 GiB of which 2.50 GiB is free. Including non-PyTorch memory, this process has 265.17 GiB memory in use. Of the allocated memory 264.14 GiB is allocated by PyTorch, and 218.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.029089792,"vramAllocatedPct":98.60667537667075,"vramReservedGb":148.55176192,"vramReservedPct":98.95484316535708,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.06 GiB is free. Including non-PyTorch memory, this process has 138.74 GiB memory in use. Of the allocated memory 137.86 GiB is allocated by PyTorch, and 158.46 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"2b","modelLabel":"2B","parameters":2008824320,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.029089792,"vramAllocatedPct":98.60667537667075,"vramReservedGb":148.55176192,"vramReservedPct":98.95484316535708,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 3.91 GiB. GPU 0 has a total capacity of 139.81 GiB of which 1.06 GiB is free. Including non-PyTorch memory, this process has 138.74 GiB memory in use. Of the allocated memory 137.86 GiB is allocated by PyTorch, and 158.46 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":4332.34105055271,"meanTps":4329.195578655956,"stepMs":472.7236328125,"jitter":0.0026054741937616326,"achievedTflops":106.73264212821864,"nominalPeakTflops":312.0,"mfuNominalPct":34.20918016930085,"configuredPeakTflops":312.0,"mfuConfiguredPct":34.20918016930085,"vramAllocatedGb":68.34494464,"vramAllocatedPct":80.43019254529524,"vramReservedGb":68.794974208,"vramReservedPct":80.95979959956934,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":5133.884228606685,"meanTps":5135.902442314099,"stepMs":797.8364562988281,"jitter":0.0014791120471413454,"achievedTflops":126.47966115910391,"nominalPeakTflops":312.0,"mfuNominalPct":40.53835293561023,"configuredPeakTflops":312.0,"mfuConfiguredPct":40.53835293561023,"vramAllocatedGb":80.674746368,"vramAllocatedPct":94.94023907839248,"vramReservedGb":81.166073856,"vramReservedPct":95.5184466498638,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.113888768,"vramAllocatedPct":98.98751553577654,"vramReservedGb":84.360036352,"vramReservedPct":99.27718871760383,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 60.75 MiB is free. Process 764857 has 79.07 GiB memory in use. Of the allocated memory 78.34 GiB is allocated by PyTorch, and 234.74 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":4335.894392855804,"meanTps":4335.789580434123,"stepMs":472.3362274169922,"jitter":0.0008970787846510358,"achievedTflops":106.82018316156983,"nominalPeakTflops":312.0,"mfuNominalPct":34.23723819281084,"configuredPeakTflops":312.0,"mfuConfiguredPct":34.23723819281084,"vramAllocatedGb":68.34494464,"vramAllocatedPct":80.43019254529524,"vramReservedGb":68.794974208,"vramReservedPct":80.95979959956934,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":5144.783252116931,"meanTps":5145.438026186683,"stepMs":796.1462707519531,"jitter":0.0013418808966732406,"achievedTflops":126.74817223943951,"nominalPeakTflops":312.0,"mfuNominalPct":40.624414179307536,"configuredPeakTflops":312.0,"mfuConfiguredPct":40.624414179307536,"vramAllocatedGb":80.674746368,"vramAllocatedPct":94.94023907839248,"vramReservedGb":81.166073856,"vramReservedPct":95.5184466498638,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.113888768,"vramAllocatedPct":98.98751553577654,"vramReservedGb":84.360036352,"vramReservedPct":99.27718871760383,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 60.75 MiB is free. Process 852916 has 79.07 GiB memory in use. Of the allocated memory 78.34 GiB is allocated by PyTorch, and 234.74 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":23221.406720818282,"meanTps":23233.867676183938,"stepMs":352.77793884277344,"jitter":0.001553290358355261,"achievedTflops":572.0884077052775,"nominalPeakTflops":2250.0,"mfuNominalPct":25.42615145356789,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.42615145356789,"vramAllocatedGb":105.334349824,"vramAllocatedPct":55.00401850858149,"vramReservedGb":106.107502592,"vramReservedPct":55.40774729441526,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25838.33116438092,"meanTps":25834.932027541676,"stepMs":634.0966796875,"jitter":0.0017711500388871348,"achievedTflops":636.5596155008237,"nominalPeakTflops":2250.0,"mfuNominalPct":28.291538466703276,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.291538466703276,"vramAllocatedGb":154.653556736,"vramAllocatedPct":80.75776905955435,"vramReservedGb":156.260892672,"vramReservedPct":81.59709579125179,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.153170432,"vramAllocatedPct":99.2951351898324,"vramReservedGb":190.423498752,"vramReservedPct":99.43629658629536,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 215.81 MiB is free. Including non-PyTorch memory, this process has 178.13 GiB memory in use. Of the allocated memory 177.09 GiB is allocated by PyTorch, and 217.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":23296.37743616874,"meanTps":23281.914217200556,"stepMs":351.6426544189453,"jitter":0.0020273021606662333,"achievedTflops":573.9354050764969,"nominalPeakTflops":2250.0,"mfuNominalPct":25.50824022562208,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.50824022562208,"vramAllocatedGb":105.334349824,"vramAllocatedPct":55.00401850858149,"vramReservedGb":106.107502592,"vramReservedPct":55.40774729441526,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25886.393607715363,"meanTps":25858.742541968786,"stepMs":632.9193725585938,"jitter":0.003178158800534362,"achievedTflops":637.7436939250207,"nominalPeakTflops":2250.0,"mfuNominalPct":28.344164174445364,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.344164174445364,"vramAllocatedGb":154.653556736,"vramAllocatedPct":80.75776905955435,"vramReservedGb":156.260892672,"vramReservedPct":81.59709579125179,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.153170432,"vramAllocatedPct":99.2951351898324,"vramReservedGb":190.423498752,"vramReservedPct":99.43629658629536,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 215.81 MiB is free. Including non-PyTorch memory, this process has 178.13 GiB memory in use. Of the allocated memory 177.09 GiB is allocated by PyTorch, and 217.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21788.204016982003,"meanTps":21787.20592607748,"stepMs":751.9665222167969,"jitter":0.0007797350282598456,"achievedTflops":536.7796659647727,"nominalPeakTflops":2250.0,"mfuNominalPct":23.856874042878783,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.856874042878783,"vramAllocatedGb":154.653556736,"vramAllocatedPct":53.805896421028436,"vramReservedGb":156.166520832,"vramReservedPct":54.33227545269258,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":25638.98873279942,"meanTps":25639.14962315046,"stepMs":1278.0535278320312,"jitter":0.0009627866281680539,"achievedTflops":631.6485652943209,"nominalPeakTflops":2250.0,"mfuNominalPct":28.073269568636483,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.073269568636483,"vramAllocatedGb":253.291970048,"vramAllocatedPct":88.12342756491213,"vramReservedGb":256.016121856,"vramReservedPct":89.07119386987088,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.267743744,"vramAllocatedPct":99.59610948272304,"vramReservedGb":286.40804864,"vramReservedPct":99.64492330156347,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 117.62 MiB is free. Including non-PyTorch memory, this process has 267.56 GiB memory in use. Of the allocated memory 266.61 GiB is allocated by PyTorch, and 133.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21771.557554460534,"meanTps":21773.49056010657,"stepMs":752.5414733886719,"jitter":0.0008961394613362377,"achievedTflops":536.3695595335678,"nominalPeakTflops":2250.0,"mfuNominalPct":23.838647090380793,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.838647090380793,"vramAllocatedGb":154.653556736,"vramAllocatedPct":53.805896421028436,"vramReservedGb":156.166520832,"vramReservedPct":54.33227545269258,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":64,"tokensPerStep":32768,"status":"complete","stable":true,"tps":25618.406779926296,"meanTps":25617.77502861005,"stepMs":1279.080322265625,"jitter":0.0007754546714300684,"achievedTflops":631.1415031344693,"nominalPeakTflops":2250.0,"mfuNominalPct":28.05073347264308,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.05073347264308,"vramAllocatedGb":253.291970048,"vramAllocatedPct":88.12342756491213,"vramReservedGb":256.016121856,"vramReservedPct":89.07119386987088,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":128,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.267743744,"vramAllocatedPct":99.59610948272304,"vramReservedGb":286.40804864,"vramReservedPct":99.64492330156347,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 117.62 MiB is free. Including non-PyTorch memory, this process has 267.56 GiB memory in use. Of the allocated memory 266.61 GiB is allocated by PyTorch, and 133.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":9807.80684533955,"meanTps":9806.857604004574,"stepMs":208.81324768066406,"jitter":0.000575823368799729,"achievedTflops":241.6275925351651,"nominalPeakTflops":989.5,"mfuNominalPct":24.419160438116734,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.419160438116734,"vramAllocatedGb":68.395014144,"vramAllocatedPct":80.44816581281836,"vramReservedGb":68.836917248,"vramReservedPct":80.96794484391766,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":12471.250850374467,"meanTps":12473.137959195317,"stepMs":328.4353790283203,"jitter":0.0005807974887593539,"achievedTflops":307.2448679298789,"nominalPeakTflops":989.5,"mfuNominalPct":31.050517223838195,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.050517223838195,"vramAllocatedGb":80.724815872,"vramAllocatedPct":94.95083017026603,"vramReservedGb":81.249959936,"vramReservedPct":95.56851959200286,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.975840256,"vramAllocatedPct":98.77477774859243,"vramReservedGb":84.213235712,"vramReservedPct":99.05400905289902,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 14.19 MiB is free. Including non-PyTorch memory, this process has 79.16 GiB memory in use. Of the allocated memory 78.19 GiB is allocated by PyTorch, and 240.40 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":9810.098652290948,"meanTps":9810.028885842008,"stepMs":208.76446533203125,"jitter":0.0004661303863264995,"achievedTflops":241.68405406677493,"nominalPeakTflops":989.5,"mfuNominalPct":24.42486650497978,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.42486650497978,"vramAllocatedGb":68.395014144,"vramAllocatedPct":80.44816581281836,"vramReservedGb":68.836917248,"vramReservedPct":80.96794484391766,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":12481.2883749505,"meanTps":12480.800564199015,"stepMs":328.17124938964844,"jitter":0.0005773356164316683,"achievedTflops":307.49215490611783,"nominalPeakTflops":989.5,"mfuNominalPct":31.075508328056376,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.075508328056376,"vramAllocatedGb":80.724815872,"vramAllocatedPct":94.95083017026603,"vramReservedGb":81.249959936,"vramReservedPct":95.56851959200286,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.975840256,"vramAllocatedPct":98.77477774859243,"vramReservedGb":84.213235712,"vramReservedPct":99.05400905289902,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 14.19 MiB is free. Including non-PyTorch memory, this process has 79.16 GiB memory in use. Of the allocated memory 78.19 GiB is allocated by PyTorch, and 240.40 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":13438.275674442,"meanTps":13435.398040214524,"stepMs":304.80101013183594,"jitter":0.0034115008135741096,"achievedTflops":331.0687343503529,"nominalPeakTflops":989.5,"mfuNominalPct":33.458184370930056,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.458184370930056,"vramAllocatedGb":80.724815872,"vramAllocatedPct":53.77325311340264,"vramReservedGb":81.22269696,"vramReservedPct":54.10490683693484,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":15247.518260240306,"meanTps":15269.830186022093,"stepMs":537.2677612304688,"jitter":0.004248586354364821,"achievedTflops":375.64168906002595,"nominalPeakTflops":989.5,"mfuNominalPct":37.96277807579848,"configuredPeakTflops":989.5,"mfuConfiguredPct":37.96277807579848,"vramAllocatedGb":105.384419328,"vramAllocatedPct":70.19976439115172,"vramReservedGb":106.308829184,"vramReservedPct":70.81554188943716,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.657158656,"vramAllocatedPct":99.02505113425775,"vramReservedGb":149.004746752,"vramReservedPct":99.2565901283516,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 395.44 MiB is free. Including non-PyTorch memory, this process has 139.42 GiB memory in use. Of the allocated memory 138.45 GiB is allocated by PyTorch, and 251.49 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":13418.553018203284,"meanTps":13402.17500052935,"stepMs":305.24900817871094,"jitter":0.003236702843152785,"achievedTflops":330.58284203818687,"nominalPeakTflops":989.5,"mfuNominalPct":33.409079538977956,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.409079538977956,"vramAllocatedGb":80.724815872,"vramAllocatedPct":53.77325311340264,"vramReservedGb":81.22269696,"vramReservedPct":54.10490683693484,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":15239.733571520177,"meanTps":15256.220061207328,"stepMs":537.5422058105469,"jitter":0.0033335848162812888,"achievedTflops":375.4499035137013,"nominalPeakTflops":989.5,"mfuNominalPct":37.94339600946956,"configuredPeakTflops":989.5,"mfuConfiguredPct":37.94339600946956,"vramAllocatedGb":105.384419328,"vramAllocatedPct":70.19976439115172,"vramReservedGb":106.308829184,"vramReservedPct":70.81554188943716,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.657158656,"vramAllocatedPct":99.02505113425775,"vramReservedGb":149.004746752,"vramReservedPct":99.2565901283516,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 395.44 MiB is free. Including non-PyTorch memory, this process has 139.42 GiB memory in use. Of the allocated memory 138.45 GiB is allocated by PyTorch, and 251.49 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":5761.900767416436,"meanTps":5762.129753552561,"stepMs":355.4382629394531,"jitter":0.0002253750386043583,"achievedTflops":141.95163432678243,"nominalPeakTflops":468.0,"mfuNominalPct":30.33154579632103,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.33154579632103,"vramAllocatedGb":68.34494464,"vramAllocatedPct":67.0219642654012,"vramReservedGb":68.774002688,"vramReservedPct":67.44271686549926,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7388.605093951981,"meanTps":7388.085563031301,"stepMs":554.3671569824219,"jitter":0.0001737609687989372,"achievedTflops":182.02753063932988,"nominalPeakTflops":468.0,"mfuNominalPct":38.89477150413032,"configuredPeakTflops":468.0,"mfuConfiguredPct":38.89477150413032,"vramAllocatedGb":80.674746368,"vramAllocatedPct":79.11309309967423,"vramReservedGb":81.166073856,"vramReservedPct":79.59490976536617,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":101.080840704,"vramAllocatedPct":99.1241785221263,"vramReservedGb":101.267275776,"vramReservedPct":99.30700469601844,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 304.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 3.75 MiB is free. Including non-PyTorch memory, this process has 94.96 GiB memory in use. Of the allocated memory 94.14 GiB is allocated by PyTorch, and 177.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":5737.636876077001,"meanTps":5735.820760421767,"stepMs":356.9413757324219,"jitter":0.00047787962899235696,"achievedTflops":141.3538630062075,"nominalPeakTflops":468.0,"mfuNominalPct":30.203816881668267,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.203816881668267,"vramAllocatedGb":68.638808064,"vramAllocatedPct":67.31013925780158,"vramReservedGb":77.397491712,"vramReservedPct":75.8992775702298,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7375.683287128747,"meanTps":7375.372341166783,"stepMs":555.33837890625,"jitter":0.0004846043770821738,"achievedTflops":181.70918576130182,"nominalPeakTflops":468.0,"mfuNominalPct":38.82674909429526,"configuredPeakTflops":468.0,"mfuConfiguredPct":38.82674909429526,"vramAllocatedGb":80.97169408,"vramAllocatedPct":79.40429267628065,"vramReservedGb":89.833603072,"vramReservedPct":88.09465815852067,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":512,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.870440448,"vramAllocatedPct":98.917851067867,"vramReservedGb":101.246304256,"vramReservedPct":99.28643913010227,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 51.75 MiB is free. Including non-PyTorch memory, this process has 94.91 GiB memory in use. Of the allocated memory 93.93 GiB is allocated by PyTorch, and 344.46 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":4276.478812256885,"meanTps":4276.149133617353,"stepMs":478.89866638183594,"jitter":0.0021566952639919647,"achievedTflops":108.08729174219164,"nominalPeakTflops":312.0,"mfuNominalPct":34.64336273788194,"configuredPeakTflops":312.0,"mfuConfiguredPct":34.64336273788194,"vramAllocatedGb":68.349385728,"vramAllocatedPct":80.43541893862736,"vramReservedGb":68.79707136,"vramReservedPct":80.96226758516865,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":5070.8392978034335,"meanTps":5070.074530639499,"stepMs":807.7558288574219,"jitter":0.0014093037157353165,"achievedTflops":128.1646210869912,"nominalPeakTflops":312.0,"mfuNominalPct":41.07840419454846,"configuredPeakTflops":312.0,"mfuConfiguredPct":41.07840419454846,"vramAllocatedGb":80.679827456,"vramAllocatedPct":94.94621864115798,"vramReservedGb":81.17026816,"vramReservedPct":95.52338262106241,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.117689856,"vramAllocatedPct":98.99198875967528,"vramReservedGb":84.364230656,"vramReservedPct":99.28212468880244,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 56.75 MiB is free. Process 764857 has 79.07 GiB memory in use. Of the allocated memory 78.34 GiB is allocated by PyTorch, and 235.12 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":4283.6376569720305,"meanTps":4283.802896868773,"stepMs":478.09832763671875,"jitter":0.0008747214758584751,"achievedTflops":108.268230353426,"nominalPeakTflops":312.0,"mfuNominalPct":34.70135588250834,"configuredPeakTflops":312.0,"mfuConfiguredPct":34.70135588250834,"vramAllocatedGb":68.349385728,"vramAllocatedPct":80.43541893862736,"vramReservedGb":68.79707136,"vramReservedPct":80.96226758516865,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":5070.308677170641,"meanTps":5070.133749952114,"stepMs":807.8403625488281,"jitter":0.001708797973111496,"achievedTflops":128.15120973862287,"nominalPeakTflops":312.0,"mfuNominalPct":41.07410568545605,"configuredPeakTflops":312.0,"mfuConfiguredPct":41.07410568545605,"vramAllocatedGb":80.679827456,"vramAllocatedPct":94.94621864115798,"vramReservedGb":81.17026816,"vramReservedPct":95.52338262106241,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.117689856,"vramAllocatedPct":98.99198875967528,"vramReservedGb":84.364230656,"vramReservedPct":99.28212468880244,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 56.75 MiB is free. Process 852916 has 79.07 GiB memory in use. Of the allocated memory 78.34 GiB is allocated by PyTorch, and 235.12 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":22546.443696896553,"meanTps":22505.08346285706,"stepMs":363.3388977050781,"jitter":0.0033139021537996866,"achievedTflops":569.8576199256912,"nominalPeakTflops":2250.0,"mfuNominalPct":25.32700533003072,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.32700533003072,"vramAllocatedGb":105.340710912,"vramAllocatedPct":55.00734017338192,"vramReservedGb":106.111696896,"vramReservedPct":55.409937497091136,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25195.159826874675,"meanTps":25160.980982929177,"stepMs":650.2836303710938,"jitter":0.0011651513053902354,"achievedTflops":636.8034802121134,"nominalPeakTflops":2250.0,"mfuNominalPct":28.30237689831615,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.30237689831615,"vramAllocatedGb":154.662477824,"vramAllocatedPct":80.7624275179802,"vramReservedGb":156.288155648,"vramReservedPct":81.611332108645,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.15697152,"vramAllocatedPct":99.2971200610074,"vramReservedGb":190.429790208,"vramReservedPct":99.43958189030917,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 209.81 MiB is free. Including non-PyTorch memory, this process has 178.13 GiB memory in use. Of the allocated memory 177.10 GiB is allocated by PyTorch, and 220.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":22601.47003816862,"meanTps":22602.835982321678,"stepMs":362.4542999267578,"jitter":0.0011327122033712072,"achievedTflops":571.248401562568,"nominalPeakTflops":2250.0,"mfuNominalPct":25.388817847225244,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.388817847225244,"vramAllocatedGb":105.340710912,"vramAllocatedPct":55.00734017338192,"vramReservedGb":106.111696896,"vramReservedPct":55.409937497091136,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":25187.5297473149,"meanTps":25152.05293661654,"stepMs":650.4806213378906,"jitter":0.0037036222606021715,"achievedTflops":636.6106312184436,"nominalPeakTflops":2250.0,"mfuNominalPct":28.293805831930825,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.293805831930825,"vramAllocatedGb":154.662477824,"vramAllocatedPct":80.7624275179802,"vramReservedGb":156.288155648,"vramReservedPct":81.611332108645,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.15697152,"vramAllocatedPct":99.2971200610074,"vramReservedGb":190.429790208,"vramReservedPct":99.43958189030917,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 209.81 MiB is free. Including non-PyTorch memory, this process has 178.13 GiB memory in use. Of the allocated memory 177.10 GiB is allocated by PyTorch, and 220.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21335.961559646723,"meanTps":21333.810181149915,"stepMs":767.9053955078125,"jitter":0.0009669441432493557,"achievedTflops":539.2628849435752,"nominalPeakTflops":2250.0,"mfuNominalPct":23.967239330825564,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.967239330825564,"vramAllocatedGb":154.662477824,"vramAllocatedPct":53.809000178530184,"vramReservedGb":156.170715136,"vramReservedPct":54.33373470323518,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":25050.09763721714,"meanTps":25044.12240024753,"stepMs":1308.0986938476562,"jitter":0.0005532107879058164,"achievedTflops":633.1370574604476,"nominalPeakTflops":2250.0,"mfuNominalPct":28.139424776019894,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.139424776019894,"vramAllocatedGb":253.306011648,"vramAllocatedPct":88.12831281224257,"vramReservedGb":256.04128768,"vramReservedPct":89.07994937312648,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.271548928,"vramAllocatedPct":99.59743335357663,"vramReservedGb":286.412242944,"vramReservedPct":99.64638255210608,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 113.62 MiB is free. Including non-PyTorch memory, this process has 267.56 GiB memory in use. Of the allocated memory 266.61 GiB is allocated by PyTorch, and 134.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21333.814843767137,"meanTps":21328.436810029747,"stepMs":767.982666015625,"jitter":0.0012853978834946086,"achievedTflops":539.2086270562452,"nominalPeakTflops":2250.0,"mfuNominalPct":23.964827869166452,"configuredPeakTflops":2250.0,"mfuConfiguredPct":23.964827869166452,"vramAllocatedGb":154.662477824,"vramAllocatedPct":53.809000178530184,"vramReservedGb":156.170715136,"vramReservedPct":54.33373470323518,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":32,"tokensPerStep":32768,"status":"complete","stable":true,"tps":25036.26990943645,"meanTps":25036.898758810457,"stepMs":1308.8211669921875,"jitter":0.0008032581091550973,"achievedTflops":632.7875639372998,"nominalPeakTflops":2250.0,"mfuNominalPct":28.123891730546656,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.123891730546656,"vramAllocatedGb":253.306011648,"vramAllocatedPct":88.12831281224257,"vramReservedGb":256.04128768,"vramReservedPct":89.07994937312648,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.271548928,"vramAllocatedPct":99.59743335357663,"vramReservedGb":286.412242944,"vramReservedPct":99.64638255210608,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 113.62 MiB is free. Including non-PyTorch memory, this process has 267.56 GiB memory in use. Of the allocated memory 266.61 GiB is allocated by PyTorch, and 134.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":9669.275429275873,"meanTps":9668.507846719069,"stepMs":211.8049087524414,"jitter":0.0006338582953136179,"achievedTflops":244.3893306016844,"nominalPeakTflops":989.5,"mfuNominalPct":24.69826484099893,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.69826484099893,"vramAllocatedGb":68.399455232,"vramAllocatedPct":80.45338954715463,"vramReservedGb":68.8390144,"vramReservedPct":80.97041157389707,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":12301.172856308835,"meanTps":12299.563346196928,"stepMs":332.97637939453125,"jitter":0.0009480458840964643,"achievedTflops":310.91010096441784,"nominalPeakTflops":989.5,"mfuNominalPct":31.420929859971483,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.420929859971483,"vramAllocatedGb":80.72989696,"vramAllocatedPct":94.9568066908509,"vramReservedGb":81.25415424,"vramReservedPct":95.57345305196166,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.97964544,"vramAllocatedPct":98.77925351451208,"vramReservedGb":84.217430016,"vramReservedPct":99.05894251285783,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 10.19 MiB is free. Including non-PyTorch memory, this process has 79.16 GiB memory in use. Of the allocated memory 78.20 GiB is allocated by PyTorch, and 240.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":9677.138321828048,"meanTps":9675.437289302288,"stepMs":211.6328125,"jitter":0.000598884298342962,"achievedTflops":244.58806390507144,"nominalPeakTflops":989.5,"mfuNominalPct":24.718349055590846,"configuredPeakTflops":989.5,"mfuConfiguredPct":24.718349055590846,"vramAllocatedGb":68.399455232,"vramAllocatedPct":80.45338954715463,"vramReservedGb":68.8390144,"vramReservedPct":80.97041157389707,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":12283.427077425613,"meanTps":12285.110910677087,"stepMs":333.4574279785156,"jitter":0.0010541411966413015,"achievedTflops":310.4615793503633,"nominalPeakTflops":989.5,"mfuNominalPct":31.37560175344753,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.37560175344753,"vramAllocatedGb":80.72989696,"vramAllocatedPct":94.9568066908509,"vramReservedGb":81.25415424,"vramReservedPct":95.57345305196166,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.97964544,"vramAllocatedPct":98.77925351451208,"vramReservedGb":84.217430016,"vramReservedPct":99.05894251285783,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 10.19 MiB is free. Including non-PyTorch memory, this process has 79.16 GiB memory in use. Of the allocated memory 78.20 GiB is allocated by PyTorch, and 240.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":13230.677494213945,"meanTps":13230.242156466218,"stepMs":309.5835418701172,"jitter":0.003163982237154987,"achievedTflops":334.40317631533924,"nominalPeakTflops":989.5,"mfuNominalPct":33.79516688381397,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.79516688381397,"vramAllocatedGb":80.72989696,"vramAllocatedPct":53.776637780535836,"vramReservedGb":81.247862784,"vramReservedPct":54.121670557101204,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":15000.66728505511,"meanTps":15007.063793109308,"stepMs":546.1090393066406,"jitter":0.00568254436277686,"achievedTflops":379.1393743189452,"nominalPeakTflops":989.5,"mfuNominalPct":38.31625814238961,"configuredPeakTflops":989.5,"mfuConfiguredPct":38.31625814238961,"vramAllocatedGb":105.390780416,"vramAllocatedPct":70.20400170518467,"vramReservedGb":106.313023488,"vramReservedPct":70.81833584279822,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.660959744,"vramAllocatedPct":99.02758315449121,"vramReservedGb":149.011038208,"vramReservedPct":99.26078105839319,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 389.44 MiB is free. Including non-PyTorch memory, this process has 139.42 GiB memory in use. Of the allocated memory 138.45 GiB is allocated by PyTorch, and 253.86 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":13208.297001782934,"meanTps":13195.770942392985,"stepMs":310.1081085205078,"jitter":0.005471321139906545,"achievedTflops":333.8375130860976,"nominalPeakTflops":989.5,"mfuNominalPct":33.73800031188455,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.73800031188455,"vramAllocatedGb":80.72989696,"vramAllocatedPct":53.776637780535836,"vramReservedGb":81.247862784,"vramReservedPct":54.121670557101204,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14992.797593694562,"meanTps":14993.04889731894,"stepMs":546.3956909179688,"jitter":0.0027385086840061943,"achievedTflops":378.94046917680566,"nominalPeakTflops":989.5,"mfuNominalPct":38.29615656157712,"configuredPeakTflops":989.5,"mfuConfiguredPct":38.29615656157712,"vramAllocatedGb":105.390780416,"vramAllocatedPct":70.20400170518467,"vramReservedGb":106.313023488,"vramReservedPct":70.81833584279822,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.660959744,"vramAllocatedPct":99.02758315449121,"vramReservedGb":149.011038208,"vramReservedPct":99.26078105839319,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 389.44 MiB is free. Including non-PyTorch memory, this process has 139.42 GiB memory in use. Of the allocated memory 138.45 GiB is allocated by PyTorch, and 253.86 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":5720.6467398218865,"meanTps":5719.976418228811,"stepMs":358.00148010253906,"jitter":0.00038022376197069423,"achievedTflops":144.58839626398762,"nominalPeakTflops":468.0,"mfuNominalPct":30.89495646666402,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.89495646666402,"vramAllocatedGb":68.349385728,"vramAllocatedPct":67.02631938548804,"vramReservedGb":68.77609984,"vramReservedPct":67.44477342209089,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7344.552342449518,"meanTps":7344.629990845925,"stepMs":557.6922607421875,"jitter":0.0001458710572022374,"achievedTflops":185.63234067215853,"nominalPeakTflops":468.0,"mfuNominalPct":39.66503005815353,"configuredPeakTflops":468.0,"mfuConfiguredPct":39.66503005815353,"vramAllocatedGb":80.679827456,"vramAllocatedPct":79.11807583102559,"vramReservedGb":81.17026816,"vramReservedPct":79.5990228785494,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":101.025921536,"vramAllocatedPct":99.07032244638332,"vramReservedGb":101.20855552,"vramReservedPct":99.24942111145316,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 59.75 MiB is free. Including non-PyTorch memory, this process has 94.90 GiB memory in use. Of the allocated memory 94.09 GiB is allocated by PyTorch, and 174.17 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":5689.352371862569,"meanTps":5689.134993285037,"stepMs":359.9706726074219,"jitter":0.0006595677138072517,"achievedTflops":143.7974363111845,"nominalPeakTflops":468.0,"mfuNominalPct":30.72594792974028,"configuredPeakTflops":468.0,"mfuConfiguredPct":30.72594792974028,"vramAllocatedGb":68.643249152,"vramAllocatedPct":67.31449437788842,"vramReservedGb":77.399588864,"vramReservedPct":75.90133412682142,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7329.48064000304,"meanTps":7329.507377119509,"stepMs":558.8390502929688,"jitter":0.0002955154418093715,"achievedTflops":185.25140589593215,"nominalPeakTflops":468.0,"mfuNominalPct":39.58363373844704,"configuredPeakTflops":468.0,"mfuConfiguredPct":39.58363373844704,"vramAllocatedGb":80.975495168,"vramAllocatedPct":79.40802018510294,"vramReservedGb":89.839894528,"vramReservedPct":88.10082782829552,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":1024,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.874245632,"vramAllocatedPct":98.92158259340141,"vramReservedGb":101.254692864,"vramReservedPct":99.29466535646874,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 43.75 MiB is free. Including non-PyTorch memory, this process has 94.92 GiB memory in use. Of the allocated memory 93.93 GiB is allocated by PyTorch, and 348.83 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":4195.879552772066,"meanTps":4193.343265745225,"stepMs":488.097900390625,"jitter":0.002012922039725917,"achievedTflops":111.40899173691356,"nominalPeakTflops":312.0,"mfuNominalPct":35.708010172087675,"configuredPeakTflops":312.0,"mfuConfiguredPct":35.708010172087675,"vramAllocatedGb":68.35729152,"vramAllocatedPct":80.44472269000404,"vramReservedGb":68.784488448,"vramReservedPct":80.94745967157282,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":4952.541692808291,"meanTps":4953.476465613529,"stepMs":827.0500793457031,"jitter":0.0012591344373631796,"achievedTflops":131.4998844917444,"nominalPeakTflops":312.0,"mfuNominalPct":42.147398875559105,"configuredPeakTflops":312.0,"mfuConfiguredPct":42.147398875559105,"vramAllocatedGb":80.688069632,"vramAllocatedPct":94.95591825838883,"vramReservedGb":81.197531136,"vramReservedPct":95.55546643385335,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.066539008,"vramAllocatedPct":98.93179304841725,"vramReservedGb":84.393590784,"vramReservedPct":99.3166764871927,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 88.75 MiB is free. Process 703773 has 79.04 GiB memory in use. Of the allocated memory 78.24 GiB is allocated by PyTorch, and 307.90 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":4194.352628275784,"meanTps":4195.324617596346,"stepMs":488.2755889892578,"jitter":0.002825775698837008,"achievedTflops":111.36844883847006,"nominalPeakTflops":312.0,"mfuNominalPct":35.69501565335579,"configuredPeakTflops":312.0,"mfuConfiguredPct":35.69501565335579,"vramAllocatedGb":68.35729152,"vramAllocatedPct":80.44472269000404,"vramReservedGb":68.784488448,"vramReservedPct":80.94745967157282,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":4958.9500828630125,"meanTps":4959.158136326284,"stepMs":825.9812927246094,"jitter":0.0017832334771374378,"achievedTflops":131.67004006119626,"nominalPeakTflops":312.0,"mfuNominalPct":42.201935917050086,"configuredPeakTflops":312.0,"mfuConfiguredPct":42.201935917050086,"vramAllocatedGb":80.688069632,"vramAllocatedPct":94.95591825838883,"vramReservedGb":81.197531136,"vramReservedPct":95.55546643385335,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.066539008,"vramAllocatedPct":98.93179304841725,"vramReservedGb":84.393590784,"vramReservedPct":99.3166764871927,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 88.75 MiB is free. Process 736416 has 79.04 GiB memory in use. Of the allocated memory 78.24 GiB is allocated by PyTorch, and 307.90 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":11185.254348422488,"meanTps":10936.28268570919,"stepMs":183.09820556640625,"jitter":0.04328745956922958,"achievedTflops":296.9908677324637,"nominalPeakTflops":2250.0,"mfuNominalPct":13.199594121442832,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.199594121442832,"vramAllocatedGb":68.35729152,"vramAllocatedPct":35.695152951007216,"vramReservedGb":68.784488448,"vramReservedPct":35.91822878309602,"warmupSteps":13,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":17802.528129120517,"meanTps":17742.853737147532,"stepMs":230.07968139648438,"jitter":0.005481101219471271,"achievedTflops":472.69271776951365,"nominalPeakTflops":2250.0,"mfuNominalPct":21.008565234200606,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.008565234200606,"vramAllocatedGb":80.688069632,"vramAllocatedPct":42.134100441839166,"vramReservedGb":81.197531136,"vramReservedPct":42.400133602363226,"warmupSteps":12,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":21631.196284470516,"meanTps":21624.887247633884,"stepMs":378.71229553222656,"jitter":0.0012877245677746852,"achievedTflops":574.3515126700886,"nominalPeakTflops":2250.0,"mfuNominalPct":25.52673389644838,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.52673389644838,"vramAllocatedGb":105.349593088,"vramAllocatedPct":55.01197831254466,"vramReservedGb":106.250108928,"vramReservedPct":55.48221418539518,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":24020.53681493243,"meanTps":24021.482821194626,"stepMs":682.0830078125,"jitter":0.0015831959254953446,"achievedTflops":637.793281211571,"nominalPeakTflops":2250.0,"mfuNominalPct":28.3463680538476,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.3463680538476,"vramAllocatedGb":154.67264,"vramAllocatedPct":80.76773405395565,"vramReservedGb":156.082634752,"vramReservedPct":81.50401217752687,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.164573696,"vramAllocatedPct":99.30108980335744,"vramReservedGb":190.350098432,"vramReservedPct":99.39796803946746,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 265.81 MiB is free. Including non-PyTorch memory, this process has 178.08 GiB memory in use. Of the allocated memory 177.10 GiB is allocated by PyTorch, and 156.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":10266.131750997498,"meanTps":10179.534395141009,"stepMs":199.49091339111328,"jitter":0.031867623378952305,"achievedTflops":272.58632499622615,"nominalPeakTflops":2250.0,"mfuNominalPct":12.11494777761005,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.11494777761005,"vramAllocatedGb":68.35729152,"vramAllocatedPct":35.695152951007216,"vramReservedGb":68.784488448,"vramReservedPct":35.91822878309602,"warmupSteps":14,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":17869.14708792343,"meanTps":17849.603678096642,"stepMs":229.22190856933594,"jitter":0.0029804408210281166,"achievedTflops":474.4615843295396,"nominalPeakTflops":2250.0,"mfuNominalPct":21.087181525757316,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.087181525757316,"vramAllocatedGb":80.688069632,"vramAllocatedPct":42.134100441839166,"vramReservedGb":81.197531136,"vramReservedPct":42.400133602363226,"warmupSteps":27,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":21626.433453220383,"meanTps":21589.85617971918,"stepMs":378.7957000732422,"jitter":0.0016439254929202724,"achievedTflops":574.2250499771737,"nominalPeakTflops":2250.0,"mfuNominalPct":25.521113332318833,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.521113332318833,"vramAllocatedGb":105.349593088,"vramAllocatedPct":55.01197831254466,"vramReservedGb":106.250108928,"vramReservedPct":55.48221418539518,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":23986.723261541872,"meanTps":23985.55131107284,"stepMs":683.0445251464844,"jitter":0.001333690662091324,"achievedTflops":636.8954637592575,"nominalPeakTflops":2250.0,"mfuNominalPct":28.306465055967,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.306465055967,"vramAllocatedGb":154.67264,"vramAllocatedPct":80.76773405395565,"vramReservedGb":156.082634752,"vramReservedPct":81.50401217752687,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.164573696,"vramAllocatedPct":99.30108980335744,"vramReservedGb":190.350098432,"vramReservedPct":99.39796803946746,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 265.81 MiB is free. Including non-PyTorch memory, this process has 178.08 GiB memory in use. Of the allocated memory 177.10 GiB is allocated by PyTorch, and 156.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":6798.741436528177,"meanTps":6797.072153388054,"stepMs":301.2322235107422,"jitter":0.0010187378349881759,"achievedTflops":180.52017914174053,"nominalPeakTflops":2250.0,"mfuNominalPct":8.023119072966246,"configuredPeakTflops":2250.0,"mfuConfiguredPct":8.023119072966246,"vramAllocatedGb":68.35729152,"vramAllocatedPct":23.78235214742398,"vramReservedGb":68.784488448,"vramReservedPct":23.930979273398112,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":10971.419752478389,"meanTps":10970.47949457454,"stepMs":373.33363342285156,"jitter":0.0008442292866089968,"achievedTflops":291.31313165043923,"nominalPeakTflops":2250.0,"mfuNominalPct":12.947250295575078,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.947250295575078,"vramAllocatedGb":80.688069632,"vramAllocatedPct":28.07238325881655,"vramReservedGb":81.197531136,"vramReservedPct":28.24963125422812,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":15826.930493399526,"meanTps":15825.376850523488,"stepMs":517.5987854003906,"jitter":0.00046897591411003224,"achievedTflops":420.2366503573558,"nominalPeakTflops":2250.0,"mfuNominalPct":18.677184460326927,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.677184460326927,"vramAllocatedGb":105.349593088,"vramAllocatedPct":36.652434081206835,"vramReservedGb":106.250108928,"vramReservedPct":36.965734745188634,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":20432.306638975566,"meanTps":20427.069807846765,"stepMs":801.8673706054688,"jitter":0.0009237335717295025,"achievedTflops":542.5185954167382,"nominalPeakTflops":2250.0,"mfuNominalPct":24.111937574077256,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.111937574077256,"vramAllocatedGb":154.67264,"vramAllocatedPct":53.8125357259874,"vramReservedGb":156.082634752,"vramReservedPct":54.303090441840546,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":23827.153108959545,"meanTps":23830.485077273235,"stepMs":1375.2377319335938,"jitter":0.0009100806035178133,"achievedTflops":632.6585571495914,"nominalPeakTflops":2250.0,"mfuNominalPct":28.118158095537392,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.118158095537392,"vramAllocatedGb":253.318733824,"vramAllocatedPct":88.13273901554854,"vramReservedGb":256.074842112,"vramReservedPct":89.0916233774673,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.279159296,"vramAllocatedPct":99.60008109528381,"vramReservedGb":286.462574592,"vramReservedPct":99.6638935586173,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 65.62 MiB is free. Including non-PyTorch memory, this process has 267.61 GiB memory in use. Of the allocated memory 266.62 GiB is allocated by PyTorch, and 174.92 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":6786.724311482445,"meanTps":6783.7235449699,"stepMs":301.76560974121094,"jitter":0.0006778581520269171,"achievedTflops":180.20110044367902,"nominalPeakTflops":2250.0,"mfuNominalPct":8.008937797496845,"configuredPeakTflops":2250.0,"mfuConfiguredPct":8.008937797496845,"vramAllocatedGb":68.35729152,"vramAllocatedPct":23.78235214742398,"vramReservedGb":68.784488448,"vramReservedPct":23.930979273398112,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":10963.649194208789,"meanTps":10960.082352424948,"stepMs":373.5982360839844,"jitter":0.0006051111221658891,"achievedTflops":291.1068077912434,"nominalPeakTflops":2250.0,"mfuNominalPct":12.938080346277484,"configuredPeakTflops":2250.0,"mfuConfiguredPct":12.938080346277484,"vramAllocatedGb":80.688069632,"vramAllocatedPct":28.07238325881655,"vramReservedGb":81.197531136,"vramReservedPct":28.24963125422812,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":15817.4245633842,"meanTps":15815.585874446608,"stepMs":517.9098510742188,"jitter":0.0002247810395106438,"achievedTflops":419.98424890845587,"nominalPeakTflops":2250.0,"mfuNominalPct":18.665966618153593,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.665966618153593,"vramAllocatedGb":105.349593088,"vramAllocatedPct":36.652434081206835,"vramReservedGb":106.250108928,"vramReservedPct":36.965734745188634,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":20442.223715079133,"meanTps":20443.614602218477,"stepMs":801.4783630371094,"jitter":0.0012557267467196904,"achievedTflops":542.7819136163625,"nominalPeakTflops":2250.0,"mfuNominalPct":24.12364060517167,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.12364060517167,"vramAllocatedGb":154.67264,"vramAllocatedPct":53.8125357259874,"vramReservedGb":156.082634752,"vramReservedPct":54.303090441840546,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":16,"tokensPerStep":32768,"status":"complete","stable":true,"tps":23829.288363593303,"meanTps":23827.319346845943,"stepMs":1375.114501953125,"jitter":0.0005791582075285279,"achievedTflops":632.7152524295337,"nominalPeakTflops":2250.0,"mfuNominalPct":28.120677885757054,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.120677885757054,"vramAllocatedGb":253.318733824,"vramAllocatedPct":88.13273901554854,"vramReservedGb":256.074842112,"vramReservedPct":89.0916233774673,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.279159296,"vramAllocatedPct":99.60008109528381,"vramReservedGb":286.462574592,"vramReservedPct":99.6638935586173,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 65.62 MiB is free. Including non-PyTorch memory, this process has 267.61 GiB memory in use. Of the allocated memory 266.62 GiB is allocated by PyTorch, and 174.92 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.735544832,"vramAllocatedPct":97.91971029866157,"vramReservedGb":24.754782208,"vramReservedPct":97.99586460606092,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 26.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 7.69 MiB is free. Including non-PyTorch memory, this process has 23.51 GiB memory in use. Of the allocated memory 23.04 GiB is allocated by PyTorch, and 14.35 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":24.735544832,"vramAllocatedPct":97.91971029866157,"vramReservedGb":24.754782208,"vramReservedPct":97.99586460606092,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 26.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 7.69 MiB is free. Including non-PyTorch memory, this process has 23.51 GiB memory in use. Of the allocated memory 23.04 GiB is allocated by PyTorch, and 14.35 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.925501952,"vramAllocatedPct":97.79215586350064,"vramReservedGb":32.95674368,"vramReservedPct":97.88494703608399,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 134.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 71.88 MiB is free. Including non-PyTorch memory, this process has 31.28 GiB memory in use. Of the allocated memory 30.66 GiB is allocated by PyTorch, and 25.79 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.925501952,"vramAllocatedPct":97.79215586350064,"vramReservedGb":32.95674368,"vramReservedPct":97.88494703608399,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 134.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 71.88 MiB is free. Including non-PyTorch memory, this process has 31.28 GiB memory in use. Of the allocated memory 30.66 GiB is allocated by PyTorch, and 25.79 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":9437.145993369066,"meanTps":9435.67129234462,"stepMs":217.01476287841797,"jitter":0.0007297142486898812,"achievedTflops":250.5750955841167,"nominalPeakTflops":989.5,"mfuNominalPct":25.32340531421088,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.32340531421088,"vramAllocatedGb":68.407361024,"vramAllocatedPct":80.4626885651263,"vramReservedGb":68.826431488,"vramReservedPct":80.95561119402065,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":11877.68818328339,"meanTps":11885.66557736798,"stepMs":344.84825134277344,"jitter":0.0026209248107575338,"achievedTflops":315.3763705611642,"nominalPeakTflops":989.5,"mfuNominalPct":31.87229616585793,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.87229616585793,"vramAllocatedGb":80.738139136,"vramAllocatedPct":94.96650137327484,"vramReservedGb":81.260445696,"vramReservedPct":95.58085324189987,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.95789568,"vramAllocatedPct":98.75367082663976,"vramReservedGb":84.22162432,"vramReservedPct":99.06387597281663,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 6.19 MiB is free. Including non-PyTorch memory, this process has 79.16 GiB memory in use. Of the allocated memory 78.19 GiB is allocated by PyTorch, and 251.51 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":9421.656741656285,"meanTps":9419.981094377285,"stepMs":217.3715362548828,"jitter":0.0005717293918478054,"achievedTflops":250.16382498056942,"nominalPeakTflops":989.5,"mfuNominalPct":25.281841837349106,"configuredPeakTflops":989.5,"mfuConfiguredPct":25.281841837349106,"vramAllocatedGb":68.407361024,"vramAllocatedPct":80.4626885651263,"vramReservedGb":68.826431488,"vramReservedPct":80.95561119402065,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":11878.5827571868,"meanTps":11878.485777181915,"stepMs":344.82228088378906,"jitter":0.0020580191858023842,"achievedTflops":315.4001232869895,"nominalPeakTflops":989.5,"mfuNominalPct":31.87469664345523,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.87469664345523,"vramAllocatedGb":80.738139136,"vramAllocatedPct":94.96650137327484,"vramReservedGb":81.260445696,"vramReservedPct":95.58085324189987,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.95789568,"vramAllocatedPct":98.75367082663976,"vramReservedGb":84.22162432,"vramReservedPct":99.06387597281663,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 6.19 MiB is free. Including non-PyTorch memory, this process has 79.16 GiB memory in use. Of the allocated memory 78.19 GiB is allocated by PyTorch, and 251.51 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":10361.921678773986,"meanTps":10366.91738471259,"stepMs":197.64673614501953,"jitter":0.0008545024612682653,"achievedTflops":275.12973910950296,"nominalPeakTflops":989.5,"mfuNominalPct":27.804925630065988,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.804925630065988,"vramAllocatedGb":68.407361024,"vramAllocatedPct":45.56822210652297,"vramReservedGb":68.826431488,"vramReservedPct":45.8473776783208,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":12745.604463069265,"meanTps":12738.409244145407,"stepMs":321.3656921386719,"jitter":0.005913850875802703,"achievedTflops":338.4212831776656,"nominalPeakTflops":989.5,"mfuNominalPct":34.20124135196216,"configuredPeakTflops":989.5,"mfuConfiguredPct":34.20124135196216,"vramAllocatedGb":80.738139136,"vramAllocatedPct":53.78212814445263,"vramReservedGb":81.260445696,"vramReservedPct":54.13005241718439,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14473.570028699605,"meanTps":14475.164228004543,"stepMs":565.9971923828125,"jitter":0.0014118861056767506,"achievedTflops":384.30222399156406,"nominalPeakTflops":989.5,"mfuNominalPct":38.83802162623184,"configuredPeakTflops":989.5,"mfuConfiguredPct":38.83802162623184,"vramAllocatedGb":105.399662592,"vramAllocatedPct":70.20991839255132,"vramReservedGb":106.250108928,"vramReservedPct":70.77642654238231,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.66856192,"vramAllocatedPct":99.03264719495813,"vramReservedGb":148.784545792,"vramReservedPct":99.10990757689592,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 523.44 MiB is free. Including non-PyTorch memory, this process has 139.29 GiB memory in use. Of the allocated memory 138.46 GiB is allocated by PyTorch, and 110.61 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":10396.640887771351,"meanTps":10396.02411881244,"stepMs":196.98670196533203,"jitter":0.0008524273240462487,"achievedTflops":276.05160352902476,"nominalPeakTflops":989.5,"mfuNominalPct":27.898090301063643,"configuredPeakTflops":989.5,"mfuConfiguredPct":27.898090301063643,"vramAllocatedGb":68.407361024,"vramAllocatedPct":45.56822210652297,"vramReservedGb":68.826431488,"vramReservedPct":45.8473776783208,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":12746.109804248175,"meanTps":12762.88114776055,"stepMs":321.3529510498047,"jitter":0.0036916130611755723,"achievedTflops":338.4347009963894,"nominalPeakTflops":989.5,"mfuNominalPct":34.20259737204541,"configuredPeakTflops":989.5,"mfuConfiguredPct":34.20259737204541,"vramAllocatedGb":80.738139136,"vramAllocatedPct":53.78212814445263,"vramReservedGb":81.260445696,"vramReservedPct":54.13005241718439,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":14491.119612332777,"meanTps":14489.747148228713,"stepMs":565.3117370605469,"jitter":0.003335265436145871,"achievedTflops":384.7682005272066,"nominalPeakTflops":989.5,"mfuNominalPct":38.885113747064835,"configuredPeakTflops":989.5,"mfuConfiguredPct":38.885113747064835,"vramAllocatedGb":105.399662592,"vramAllocatedPct":70.20991839255132,"vramReservedGb":106.250108928,"vramReservedPct":70.77642654238231,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.66856192,"vramAllocatedPct":99.03264719495813,"vramReservedGb":148.784545792,"vramReservedPct":99.10990757689592,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 523.44 MiB is free. Including non-PyTorch memory, this process has 139.29 GiB memory in use. Of the allocated memory 138.46 GiB is allocated by PyTorch, and 110.61 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":49.667197952,"vramAllocatedPct":97.62240739631517,"vramReservedGb":50.197430272,"vramReservedPct":98.66459535319888,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 266.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 122.50 MiB is free. Process 2308332 has 47.25 GiB memory in use. Of the allocated memory 46.26 GiB is allocated by PyTorch, and 505.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":49.667197952,"vramAllocatedPct":97.62240739631517,"vramReservedGb":50.197430272,"vramReservedPct":98.66459535319888,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 266.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 122.50 MiB is free. Process 2312978 has 47.25 GiB memory in use. Of the allocated memory 46.26 GiB is allocated by PyTorch, and 505.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.085580288,"vramAllocatedPct":98.14172364347208,"vramReservedGb":50.616860672,"vramReservedPct":99.18275725681764,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 134.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 60.25 MiB is free. Process 1348507 has 47.46 GiB memory in use. Of the allocated memory 46.65 GiB is allocated by PyTorch, and 506.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.085580288,"vramAllocatedPct":98.14172364347208,"vramReservedGb":50.616860672,"vramReservedPct":99.18275725681764,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 134.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 60.25 MiB is free. Process 1380474 has 47.46 GiB memory in use. Of the allocated memory 46.65 GiB is allocated by PyTorch, and 506.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":5632.8718849548495,"meanTps":5631.6643696119445,"stepMs":363.580078125,"jitter":0.0003777497794153196,"achievedTflops":149.56401140529076,"nominalPeakTflops":468.0,"mfuNominalPct":31.958122095147594,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.958122095147594,"vramAllocatedGb":68.35729152,"vramAllocatedPct":67.03407214191654,"vramReservedGb":68.784488448,"vramReservedPct":67.45299964845736,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7216.373520052497,"meanTps":7216.4740902204,"stepMs":567.5981140136719,"jitter":3.7864020510719495e-05,"achievedTflops":191.60914600965066,"nominalPeakTflops":468.0,"mfuNominalPct":40.94212521573732,"configuredPeakTflops":468.0,"mfuConfiguredPct":40.94212521573732,"vramAllocatedGb":80.688069632,"vramAllocatedPct":79.12615845993474,"vramReservedGb":81.197531136,"vramReservedPct":79.62575811424043,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.894063104,"vramAllocatedPct":98.94101645068538,"vramReservedGb":101.154029568,"vramReservedPct":99.1959506400711,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 134.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 111.75 MiB is free. Including non-PyTorch memory, this process has 94.85 GiB memory in use. Of the allocated memory 93.96 GiB is allocated by PyTorch, and 247.92 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":5631.099196835254,"meanTps":5629.777508827589,"stepMs":363.6945343017578,"jitter":0.0007307540459788585,"achievedTflops":149.51694299124688,"nominalPeakTflops":468.0,"mfuNominalPct":31.948064741719417,"configuredPeakTflops":468.0,"mfuConfiguredPct":31.948064741719417,"vramAllocatedGb":68.35729152,"vramAllocatedPct":67.03407214191654,"vramReservedGb":68.805459968,"vramReservedPct":67.47356521437354,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":7218.351293557225,"meanTps":7216.733327409081,"stepMs":567.4425964355469,"jitter":0.00019894025796227688,"achievedTflops":191.6616598507356,"nominalPeakTflops":468.0,"mfuNominalPct":40.95334612195205,"configuredPeakTflops":468.0,"mfuConfiguredPct":40.95334612195205,"vramAllocatedGb":80.688069632,"vramAllocatedPct":79.12615845993474,"vramReservedGb":81.218502656,"vramReservedPct":79.64632368015661,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":2048,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":101.033523712,"vramAllocatedPct":99.07777746402793,"vramReservedGb":101.237915648,"vramReservedPct":99.2782129037358,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 53.75 MiB is free. Including non-PyTorch memory, this process has 94.91 GiB memory in use. Of the allocated memory 94.09 GiB is allocated by PyTorch, and 174.92 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":4747.232573343545,"meanTps":4747.491921265783,"stepMs":862.8184814453125,"jitter":0.0008374512553321499,"achievedTflops":138.17452082875954,"nominalPeakTflops":312.0,"mfuNominalPct":44.28670539383319,"configuredPeakTflops":312.0,"mfuConfiguredPct":44.28670539383319,"vramAllocatedGb":80.703561216,"vramAllocatedPct":94.9741491764255,"vramReservedGb":80.956358656,"vramReservedPct":95.2716480899334,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.023023104,"vramAllocatedPct":98.8805823472317,"vramReservedGb":84.401979392,"vramReservedPct":99.32654842958992,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 20.75 MiB is free. Process 764857 has 79.11 GiB memory in use. Of the allocated memory 78.25 GiB is allocated by PyTorch, and 361.40 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":4748.919303590437,"meanTps":4750.325536591834,"stepMs":862.5120239257812,"jitter":0.0012256207365768578,"achievedTflops":138.2236153569991,"nominalPeakTflops":312.0,"mfuNominalPct":44.30244081955099,"configuredPeakTflops":312.0,"mfuConfiguredPct":44.30244081955099,"vramAllocatedGb":80.703561216,"vramAllocatedPct":94.9741491764255,"vramReservedGb":80.956358656,"vramReservedPct":95.2716480899334,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.023023104,"vramAllocatedPct":98.8805823472317,"vramReservedGb":84.401979392,"vramReservedPct":99.32654842958992,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 20.75 MiB is free. Process 852916 has 79.11 GiB memory in use. Of the allocated memory 78.25 GiB is allocated by PyTorch, and 361.40 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":19620.434644131485,"meanTps":19606.246242099915,"stepMs":417.5238800048828,"jitter":0.0010100246008775769,"achievedTflops":571.0788577386747,"nominalPeakTflops":2250.0,"mfuNominalPct":25.38128256616332,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.38128256616332,"vramAllocatedGb":105.36543744,"vramAllocatedPct":55.02025199565108,"vramReservedGb":106.134765568,"vramReservedPct":55.42198361180848,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21761.752124713068,"meanTps":21753.969670300463,"stepMs":752.8805541992188,"jitter":0.001448802422661336,"achievedTflops":633.404752299433,"nominalPeakTflops":2250.0,"mfuNominalPct":28.151322324419244,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.151322324419244,"vramAllocatedGb":154.689124352,"vramAllocatedPct":80.77634193546841,"vramReservedGb":156.307030016,"vramReservedPct":81.62118802068646,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.179778048,"vramAllocatedPct":99.30902928805752,"vramReservedGb":190.446567424,"vramReservedPct":99.44834270101269,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 193.81 MiB is free. Including non-PyTorch memory, this process has 178.15 GiB memory in use. Of the allocated memory 177.12 GiB is allocated by PyTorch, and 214.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":19697.52414866056,"meanTps":19695.56382418663,"stepMs":415.88983154296875,"jitter":0.0006515534847024607,"achievedTflops":573.3226503451384,"nominalPeakTflops":2250.0,"mfuNominalPct":25.48100668200615,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.48100668200615,"vramAllocatedGb":105.36543744,"vramAllocatedPct":55.02025199565108,"vramReservedGb":106.134765568,"vramReservedPct":55.42198361180848,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":21859.80068596017,"meanTps":21849.388485952237,"stepMs":749.5036315917969,"jitter":0.0009004960436258778,"achievedTflops":636.2585861402988,"nominalPeakTflops":2250.0,"mfuNominalPct":28.27815938401328,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.27815938401328,"vramAllocatedGb":154.689124352,"vramAllocatedPct":80.77634193546841,"vramReservedGb":156.307030016,"vramReservedPct":81.62118802068646,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.179778048,"vramAllocatedPct":99.30902928805752,"vramReservedGb":190.446567424,"vramReservedPct":99.44834270101269,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 193.81 MiB is free. Including non-PyTorch memory, this process has 178.15 GiB memory in use. Of the allocated memory 177.12 GiB is allocated by PyTorch, and 214.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":18835.797363834874,"meanTps":18834.37335532351,"stepMs":869.8330993652344,"jitter":0.0003800819438417235,"achievedTflops":548.2409456384455,"nominalPeakTflops":2250.0,"mfuNominalPct":24.366264250597578,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.366264250597578,"vramAllocatedGb":154.689124352,"vramAllocatedPct":53.818270837128715,"vramReservedGb":156.193783808,"vramReservedPct":54.34176058121949,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":21693.93656169853,"meanTps":21693.197603897115,"stepMs":1510.4681396484375,"jitter":0.0007661170275312546,"achievedTflops":631.4308901008792,"nominalPeakTflops":2250.0,"mfuNominalPct":28.06359511559463,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.06359511559463,"vramAllocatedGb":253.336498176,"vramAllocatedPct":88.13891945461422,"vramReservedGb":256.064356352,"vramReservedPct":89.08797525111079,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.294380032,"vramAllocatedPct":99.60537657869818,"vramReservedGb":286.433214464,"vramReservedPct":99.65367880481908,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 93.62 MiB is free. Including non-PyTorch memory, this process has 267.58 GiB memory in use. Of the allocated memory 266.63 GiB is allocated by PyTorch, and 132.40 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":18847.504195806723,"meanTps":18848.688451138172,"stepMs":869.2928161621094,"jitter":0.00041180749705099036,"achievedTflops":548.5816885603778,"nominalPeakTflops":2250.0,"mfuNominalPct":24.38140838046123,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.38140838046123,"vramAllocatedGb":154.689124352,"vramAllocatedPct":53.818270837128715,"vramReservedGb":156.193783808,"vramReservedPct":54.34176058121949,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":8,"tokensPerStep":32768,"status":"complete","stable":true,"tps":21693.900620711924,"meanTps":21699.067263417284,"stepMs":1510.4706420898438,"jitter":0.0016433027251868276,"achievedTflops":631.4298439906402,"nominalPeakTflops":2250.0,"mfuNominalPct":28.06354862180623,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.06354862180623,"vramAllocatedGb":253.336498176,"vramAllocatedPct":88.13891945461422,"vramReservedGb":256.064356352,"vramReservedPct":89.08797525111079,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.294380032,"vramAllocatedPct":99.60537657869818,"vramReservedGb":286.433214464,"vramReservedPct":99.65367880481908,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 93.62 MiB is free. Including non-PyTorch memory, this process has 267.58 GiB memory in use. Of the allocated memory 266.63 GiB is allocated by PyTorch, and 132.40 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":11238.95903471601,"meanTps":11237.487466089846,"stepMs":364.4465637207031,"jitter":0.0008123012497294255,"achievedTflops":327.12485753403604,"nominalPeakTflops":989.5,"mfuNominalPct":33.05961167600162,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.05961167600162,"vramAllocatedGb":80.75363072,"vramAllocatedPct":94.98472301609387,"vramReservedGb":81.019273216,"vramReservedPct":95.29717929426856,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.885036032,"vramAllocatedPct":98.66797123118347,"vramReservedGb":84.21113856,"vramReservedPct":99.05154232291962,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 16.19 MiB is free. Including non-PyTorch memory, this process has 79.15 GiB memory in use. Of the allocated memory 78.12 GiB is allocated by PyTorch, and 311.00 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":11243.668460138375,"meanTps":11236.213440468337,"stepMs":364.2939147949219,"jitter":0.0006162190778176056,"achievedTflops":327.2619316274285,"nominalPeakTflops":989.5,"mfuNominalPct":33.07346454041723,"configuredPeakTflops":989.5,"mfuConfiguredPct":33.07346454041723,"vramAllocatedGb":80.75363072,"vramAllocatedPct":94.98472301609387,"vramReservedGb":81.019273216,"vramReservedPct":95.29717929426856,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.885036032,"vramAllocatedPct":98.66797123118347,"vramReservedGb":84.21113856,"vramReservedPct":99.05154232291962,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 16.19 MiB is free. Including non-PyTorch memory, this process has 79.15 GiB memory in use. Of the allocated memory 78.12 GiB is allocated by PyTorch, and 311.00 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":12059.42606960146,"meanTps":12040.898588484415,"stepMs":339.6513214111328,"jitter":0.004396302746674672,"achievedTflops":351.005642317505,"nominalPeakTflops":989.5,"mfuNominalPct":35.473031057858016,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.473031057858016,"vramAllocatedGb":80.75363072,"vramAllocatedPct":53.79244755935077,"vramReservedGb":80.998301696,"vramReservedPct":53.95543033211811,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":13563.660900312076,"meanTps":13566.828105085038,"stepMs":603.9667358398438,"jitter":0.0031972402808539855,"achievedTflops":394.78839863630475,"nominalPeakTflops":989.5,"mfuNominalPct":39.8977664109454,"configuredPeakTflops":989.5,"mfuConfiguredPct":39.8977664109454,"vramAllocatedGb":105.415506944,"vramAllocatedPct":70.22047279693503,"vramReservedGb":106.33609216,"vramReservedPct":70.83370258628405,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.683766272,"vramAllocatedPct":99.04277527589198,"vramReservedGb":149.04459264,"vramReservedPct":99.28313268528167,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 357.44 MiB is free. Including non-PyTorch memory, this process has 139.45 GiB memory in use. Of the allocated memory 138.47 GiB is allocated by PyTorch, and 264.11 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":12023.188071885406,"meanTps":12015.945667955119,"stepMs":340.67503356933594,"jitter":0.002796134500204891,"achievedTflops":349.9508871748297,"nominalPeakTflops":989.5,"mfuNominalPct":35.36643629861846,"configuredPeakTflops":989.5,"mfuConfiguredPct":35.36643629861846,"vramAllocatedGb":80.75363072,"vramAllocatedPct":53.79244755935077,"vramReservedGb":80.998301696,"vramReservedPct":53.95543033211811,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":13574.05743555804,"meanTps":13564.52251160411,"stepMs":603.504150390625,"jitter":0.003202764777485656,"achievedTflops":395.09100362851785,"nominalPeakTflops":989.5,"mfuNominalPct":39.92834801703061,"configuredPeakTflops":989.5,"mfuConfiguredPct":39.92834801703061,"vramAllocatedGb":105.415506944,"vramAllocatedPct":70.22047279693503,"vramReservedGb":106.33609216,"vramReservedPct":70.83370258628405,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.683766272,"vramAllocatedPct":99.04277527589198,"vramReservedGb":149.04459264,"vramReservedPct":99.28313268528167,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 357.44 MiB is free. Including non-PyTorch memory, this process has 139.45 GiB memory in use. Of the allocated memory 138.47 GiB is allocated by PyTorch, and 264.11 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":6996.6764963416435,"meanTps":6996.394022918296,"stepMs":585.4208068847656,"jitter":0.00023124241486072953,"achievedTflops":203.6475793716898,"nominalPeakTflops":468.0,"mfuNominalPct":43.514440036685855,"configuredPeakTflops":468.0,"mfuConfiguredPct":43.514440036685855,"vramAllocatedGb":80.703561216,"vramAllocatedPct":79.14135016715949,"vramReservedGb":80.956358656,"vramReservedPct":79.38925410620443,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.909267456,"vramAllocatedPct":98.9559264859746,"vramReservedGb":101.21904128,"vramReservedPct":99.25970389441125,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 134.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 49.75 MiB is free. Including non-PyTorch memory, this process has 94.91 GiB memory in use. Of the allocated memory 93.98 GiB is allocated by PyTorch, and 295.42 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":6987.67920418035,"meanTps":6985.813146227932,"stepMs":586.1745910644531,"jitter":0.0002347272397955223,"achievedTflops":203.38570121132247,"nominalPeakTflops":468.0,"mfuNominalPct":43.4584831648125,"configuredPeakTflops":468.0,"mfuConfiguredPct":43.4584831648125,"vramAllocatedGb":80.855662592,"vramAllocatedPct":79.29050725611006,"vramReservedGb":85.966454784,"vramReservedPct":84.30236780357828,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":4096,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.045567488,"vramAllocatedPct":98.10894550301768,"vramReservedGb":101.26098432,"vramReservedPct":99.30083502624359,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 68.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 9.75 MiB is free. Including non-PyTorch memory, this process has 94.95 GiB memory in use. Of the allocated memory 93.17 GiB is allocated by PyTorch, and 1.13 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.112152064,"vramAllocatedPct":98.98547173520211,"vramReservedGb":84.383105024,"vramReservedPct":99.30433655919617,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 98.75 MiB is free. Process 764857 has 79.03 GiB memory in use. Of the allocated memory 78.28 GiB is allocated by PyTorch, and 254.40 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.112152064,"vramAllocatedPct":98.98547173520211,"vramReservedGb":84.383105024,"vramReservedPct":99.30433655919617,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 98.75 MiB is free. Process 852916 has 79.03 GiB memory in use. Of the allocated memory 78.28 GiB is allocated by PyTorch, and 254.40 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":16707.908462962794,"meanTps":16652.772485253317,"stepMs":490.3067321777344,"jitter":0.0007724734229030314,"achievedTflops":571.6609693672902,"nominalPeakTflops":2250.0,"mfuNominalPct":25.40715419410179,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.40715419410179,"vramAllocatedGb":105.396100608,"vramAllocatedPct":55.036263842337576,"vramReservedGb":105.639837696,"vramReservedPct":55.16353969605462,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":18326.67197186968,"meanTps":18326.841198465576,"stepMs":893.99755859375,"jitter":0.0005123772807684934,"achievedTflops":627.0469513248433,"nominalPeakTflops":2250.0,"mfuNominalPct":27.868753392215257,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.868753392215257,"vramAllocatedGb":154.720173056,"vramAllocatedPct":80.7925551032749,"vramReservedGb":156.355264512,"vramReservedPct":81.64637535145908,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.210186752,"vramAllocatedPct":99.32490825745764,"vramReservedGb":190.496899072,"vramReservedPct":99.47462513312325,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 145.81 MiB is free. Including non-PyTorch memory, this process has 178.20 GiB memory in use. Of the allocated memory 177.15 GiB is allocated by PyTorch, and 233.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":16765.5670744271,"meanTps":16749.726871687155,"stepMs":488.6205139160156,"jitter":0.0009540851584558403,"achievedTflops":573.6337583489347,"nominalPeakTflops":2250.0,"mfuNominalPct":25.4948337043971,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.4948337043971,"vramAllocatedGb":105.396100608,"vramAllocatedPct":55.036263842337576,"vramReservedGb":105.639837696,"vramReservedPct":55.16353969605462,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":18386.308119369874,"meanTps":18367.032905896307,"stepMs":891.0978698730469,"jitter":0.0010693691312127144,"achievedTflops":629.0874016879073,"nominalPeakTflops":2250.0,"mfuNominalPct":27.959440075018104,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.959440075018104,"vramAllocatedGb":154.720173056,"vramAllocatedPct":80.7925551032749,"vramReservedGb":156.355264512,"vramReservedPct":81.64637535145908,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.210186752,"vramAllocatedPct":99.32490825745764,"vramReservedGb":190.496899072,"vramReservedPct":99.47462513312325,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 145.81 MiB is free. Including non-PyTorch memory, this process has 178.20 GiB memory in use. Of the allocated memory 177.15 GiB is allocated by PyTorch, and 233.43 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":16239.087650785708,"meanTps":16240.46335714434,"stepMs":1008.9236755371094,"jitter":0.00020027856278011789,"achievedTflops":555.6202686091543,"nominalPeakTflops":2250.0,"mfuNominalPct":24.694234160406857,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.694234160406857,"vramAllocatedGb":154.720173056,"vramAllocatedPct":53.82907306752477,"vramReservedGb":156.223143936,"vramReservedPct":54.351975335017706,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":18403.81221225774,"meanTps":18403.393003058114,"stepMs":1780.5006713867188,"jitter":0.0006065239410095865,"achievedTflops":629.6863041017166,"nominalPeakTflops":2250.0,"mfuNominalPct":27.986057960076295,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.986057960076295,"vramAllocatedGb":253.36818688,"vramAllocatedPct":88.14994434897247,"vramReservedGb":256.09371648,"vramReservedPct":89.09819000490901,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.324821504,"vramAllocatedPct":99.61596754552691,"vramReservedGb":286.462574592,"vramReservedPct":99.6638935586173,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 65.62 MiB is free. Including non-PyTorch memory, this process has 267.61 GiB memory in use. Of the allocated memory 266.66 GiB is allocated by PyTorch, and 131.37 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":16239.605386154362,"meanTps":16240.438400280906,"stepMs":1008.8915100097656,"jitter":0.0002823542393761621,"achievedTflops":555.6379829211148,"nominalPeakTflops":2250.0,"mfuNominalPct":24.695021463160657,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.695021463160657,"vramAllocatedGb":154.720173056,"vramAllocatedPct":53.82907306752477,"vramReservedGb":156.223143936,"vramReservedPct":54.351975335017706,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":4,"tokensPerStep":32768,"status":"complete","stable":true,"tps":18403.3321262598,"meanTps":18398.363036128096,"stepMs":1780.547119140625,"jitter":0.0008329195210116492,"achievedTflops":629.6698779627076,"nominalPeakTflops":2250.0,"mfuNominalPct":27.98532790945367,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.98532790945367,"vramAllocatedGb":253.36818688,"vramAllocatedPct":88.14994434897247,"vramReservedGb":256.09371648,"vramReservedPct":89.09819000490901,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.324821504,"vramAllocatedPct":99.61596754552691,"vramReservedGb":286.462574592,"vramReservedPct":99.6638935586173,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 65.62 MiB is free. Including non-PyTorch memory, this process has 267.61 GiB memory in use. Of the allocated memory 266.66 GiB is allocated by PyTorch, and 131.37 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.846205952,"vramAllocatedPct":98.6222981839086,"vramReservedGb":84.225818624,"vramReservedPct":99.06880943277544,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 34.19 MiB is free. Including non-PyTorch memory, this process has 79.14 GiB memory in use. Of the allocated memory 78.09 GiB is allocated by PyTorch, and 330.03 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.846205952,"vramAllocatedPct":98.6222981839086,"vramReservedGb":84.225818624,"vramReservedPct":99.06880943277544,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 34.19 MiB is free. Including non-PyTorch memory, this process has 79.14 GiB memory in use. Of the allocated memory 78.09 GiB is allocated by PyTorch, and 330.03 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":12080.061508372684,"meanTps":12071.8826312165,"stepMs":678.1422424316406,"jitter":0.0011951477238234464,"achievedTflops":413.31921869221446,"nominalPeakTflops":989.5,"mfuNominalPct":41.770512247823596,"configuredPeakTflops":989.5,"mfuConfiguredPct":41.770512247823596,"vramAllocatedGb":105.446170112,"vramAllocatedPct":70.2408984650064,"vramReservedGb":105.849552896,"vramReservedPct":70.50960399640104,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.714174976,"vramAllocatedPct":99.06303143775966,"vramReservedGb":148.889403392,"vramReservedPct":99.17975641092244,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 425.44 MiB is free. Including non-PyTorch memory, this process has 139.39 GiB memory in use. Of the allocated memory 138.50 GiB is allocated by PyTorch, and 167.11 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":12070.439427819829,"meanTps":12074.399603990412,"stepMs":678.6828308105469,"jitter":0.002714857022631198,"achievedTflops":412.98999927445374,"nominalPeakTflops":989.5,"mfuNominalPct":41.73724095749911,"configuredPeakTflops":989.5,"mfuConfiguredPct":41.73724095749911,"vramAllocatedGb":105.446170112,"vramAllocatedPct":70.2408984650064,"vramReservedGb":105.849552896,"vramReservedPct":70.50960399640104,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.714174976,"vramAllocatedPct":99.06303143775966,"vramReservedGb":148.889403392,"vramReservedPct":99.17975641092244,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 425.44 MiB is free. Including non-PyTorch memory, this process has 139.39 GiB memory in use. Of the allocated memory 138.50 GiB is allocated by PyTorch, and 167.11 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.93967616,"vramAllocatedPct":98.98574655655305,"vramReservedGb":101.193875456,"vramReservedPct":99.23502521531184,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 134.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 73.75 MiB is free. Including non-PyTorch memory, this process has 94.89 GiB memory in use. Of the allocated memory 94.01 GiB is allocated by PyTorch, and 242.42 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":99.092936192,"vramAllocatedPct":97.1747546712755,"vramReservedGb":101.256790016,"vramReservedPct":99.29672191306035,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 69.75 MiB is free. Including non-PyTorch memory, this process has 94.89 GiB memory in use. Of the allocated memory 92.29 GiB is allocated by PyTorch, and 1.96 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":13990.38980917002,"meanTps":13986.12154569475,"stepMs":1171.089599609375,"jitter":0.0008155132512769496,"achievedTflops":621.6254411103833,"nominalPeakTflops":2250.0,"mfuNominalPct":27.627797382683703,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.627797382683703,"vramAllocatedGb":154.781179392,"vramAllocatedPct":80.82441169744472,"vramReservedGb":155.554152448,"vramReservedPct":81.22804664036599,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.035992064,"vramAllocatedPct":99.233946402575,"vramReservedGb":190.51577344,"vramReservedPct":99.48448104516471,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 367.81 MiB is free. Including non-PyTorch memory, this process has 177.98 GiB memory in use. Of the allocated memory 176.77 GiB is allocated by PyTorch, and 401.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":13996.057637370835,"meanTps":13990.8263677837,"stepMs":1170.6153564453125,"jitter":0.0006014083263040762,"achievedTflops":621.8772758521974,"nominalPeakTflops":2250.0,"mfuNominalPct":27.63899003787544,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.63899003787544,"vramAllocatedGb":154.781179392,"vramAllocatedPct":80.82441169744472,"vramReservedGb":155.554152448,"vramReservedPct":81.22804664036599,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.035992064,"vramAllocatedPct":99.233946402575,"vramReservedGb":190.51577344,"vramReservedPct":99.48448104516471,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 367.81 MiB is free. Including non-PyTorch memory, this process has 177.98 GiB memory in use. Of the allocated memory 176.77 GiB is allocated by PyTorch, and 401.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":12809.985200313069,"meanTps":12810.57635875108,"stepMs":1279.0022583007812,"jitter":0.0002520151563530399,"achievedTflops":569.1773288220122,"nominalPeakTflops":2250.0,"mfuNominalPct":25.29677016986721,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.29677016986721,"vramAllocatedGb":154.781179392,"vramAllocatedPct":53.850297930794135,"vramReservedGb":155.554152448,"vramReservedPct":54.11922487347272,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":14138.09120845384,"meanTps":14136.820384843795,"stepMs":2317.7103271484375,"jitter":0.00016585532080564045,"achievedTflops":628.1881565696942,"nominalPeakTflops":2250.0,"mfuNominalPct":27.919473625319746,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.919473625319746,"vramAllocatedGb":253.429644288,"vramAllocatedPct":88.17132614580238,"vramReservedGb":256.305528832,"vramReservedPct":89.1718821573104,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.150823424,"vramAllocatedPct":99.55543144879866,"vramReservedGb":286.340939776,"vramReservedPct":99.62157529288184,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 183.62 MiB is free. Including non-PyTorch memory, this process has 267.49 GiB memory in use. Of the allocated memory 266.50 GiB is allocated by PyTorch, and 179.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":12800.68485109343,"meanTps":12800.530844768318,"stepMs":1279.9315185546875,"jitter":0.00045770959616503226,"achievedTflops":568.7640927531825,"nominalPeakTflops":2250.0,"mfuNominalPct":25.278404122363668,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.278404122363668,"vramAllocatedGb":154.781179392,"vramAllocatedPct":53.850297930794135,"vramReservedGb":155.554152448,"vramReservedPct":54.11922487347272,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":16384,"batch":2,"tokensPerStep":32768,"status":"complete","stable":true,"tps":14134.332561137657,"meanTps":14133.210714264937,"stepMs":2318.32666015625,"jitter":0.00024317418758857803,"achievedTflops":628.0211511589966,"nominalPeakTflops":2250.0,"mfuNominalPct":27.912051162622074,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.912051162622074,"vramAllocatedGb":253.429644288,"vramAllocatedPct":88.17132614580238,"vramReservedGb":256.305528832,"vramReservedPct":89.1718821573104,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":16384,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.150823424,"vramAllocatedPct":99.55543144879866,"vramReservedGb":286.340939776,"vramReservedPct":99.62157529288184,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 183.62 MiB is free. Including non-PyTorch memory, this process has 267.49 GiB memory in use. Of the allocated memory 266.50 GiB is allocated by PyTorch, and 179.31 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.774992384,"vramAllocatedPct":99.10354376149505,"vramReservedGb":149.220753408,"vramReservedPct":99.40047872644621,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 269.44 MiB is free. Including non-PyTorch memory, this process has 139.54 GiB memory in use. Of the allocated memory 138.56 GiB is allocated by PyTorch, and 265.11 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.774992384,"vramAllocatedPct":99.10354376149505,"vramReservedGb":149.220753408,"vramReservedPct":99.40047872644621,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 1000.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 269.44 MiB is free. Including non-PyTorch memory, this process has 139.54 GiB memory in use. Of the allocated memory 138.56 GiB is allocated by PyTorch, and 265.11 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.453090816,"vramAllocatedPct":99.45174911852898,"vramReservedGb":190.51577344,"vramReservedPct":99.48448104516471,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 87.81 MiB is free. Including non-PyTorch memory, this process has 178.25 GiB memory in use. Of the allocated memory 177.37 GiB is allocated by PyTorch, and 59.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.453090816,"vramAllocatedPct":99.45174911852898,"vramReservedGb":190.51577344,"vramReservedPct":99.48448104516471,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 87.75 MiB is free. Including non-PyTorch memory, this process has 178.25 GiB memory in use. Of the allocated memory 177.37 GiB is allocated by PyTorch, and 59.78 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":9700.303325713001,"meanTps":9700.288182646842,"stepMs":3378.0386962890625,"jitter":0.00032815235472489265,"achievedTflops":629.2292511404722,"nominalPeakTflops":2250.0,"mfuNominalPct":27.9657444951321,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.9657444951321,"vramAllocatedGb":253.55133696,"vramAllocatedPct":88.21366454036001,"vramReservedGb":254.166433792,"vramReservedPct":88.42766438058348,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.802827264,"vramAllocatedPct":99.43435925534217,"vramReservedGb":286.317871104,"vramReservedPct":99.61354941489753,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 203.62 MiB is free. Including non-PyTorch memory, this process has 267.47 GiB memory in use. Of the allocated memory 266.17 GiB is allocated by PyTorch, and 491.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":32768,"batch":1,"tokensPerStep":32768,"status":"complete","stable":true,"tps":9705.316806007304,"meanTps":9705.570019829867,"stepMs":3376.293701171875,"jitter":0.00030023367778800645,"achievedTflops":629.5544603989116,"nominalPeakTflops":2250.0,"mfuNominalPct":27.980198239951626,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.980198239951626,"vramAllocatedGb":253.55133696,"vramAllocatedPct":88.21366454036001,"vramReservedGb":254.166433792,"vramReservedPct":88.42766438058348,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"4b","modelLabel":"4B","parameters":3999611392,"context":32768,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.802827264,"vramAllocatedPct":99.43435925534217,"vramReservedGb":286.317871104,"vramReservedPct":99.61354941489753,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 203.62 MiB is free. Including non-PyTorch memory, this process has 267.47 GiB memory in use. Of the allocated memory 266.17 GiB is allocated by PyTorch, and 491.18 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":11579.413435237117,"meanTps":11582.008006425078,"stepMs":353.73121643066406,"jitter":0.0005952878961203261,"achievedTflops":565.1288521498745,"nominalPeakTflops":2250.0,"mfuNominalPct":25.11683787332776,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.11683787332776,"vramAllocatedGb":149.85737472,"vramAllocatedPct":78.25327470591395,"vramReservedGb":150.325952512,"vramReservedPct":78.49795900488141,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":13696.30146659646,"meanTps":13696.177147243628,"stepMs":598.11767578125,"jitter":0.0005165126189339113,"achievedTflops":668.4427643771924,"nominalPeakTflops":2250.0,"mfuNominalPct":29.708567305652995,"configuredPeakTflops":2250.0,"mfuConfiguredPct":29.708567305652995,"vramAllocatedGb":187.929870848,"vramAllocatedPct":98.13416147448893,"vramReservedGb":188.773040128,"vramReservedPct":98.57445183333652,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.296793088,"vramAllocatedPct":99.3701327878816,"vramReservedGb":190.54723072,"vramReservedPct":99.50090756523382,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 57.81 MiB is free. Including non-PyTorch memory, this process has 178.28 GiB memory in use. Of the allocated memory 177.23 GiB is allocated by PyTorch, and 238.84 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":8,"tokensPerStep":4096,"status":"complete","stable":true,"tps":11587.577978653442,"meanTps":11584.85961824695,"stepMs":353.4819793701172,"jitter":0.0010066486127499616,"achievedTflops":565.5273195744121,"nominalPeakTflops":2250.0,"mfuNominalPct":25.13454753664054,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.13454753664054,"vramAllocatedGb":149.85737472,"vramAllocatedPct":78.25327470591395,"vramReservedGb":150.325952512,"vramReservedPct":78.49795900488141,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":13687.201862110986,"meanTps":13684.501545165635,"stepMs":598.5153198242188,"jitter":0.0023674397860509937,"achievedTflops":667.9986616541438,"nominalPeakTflops":2250.0,"mfuNominalPct":29.688829406850836,"configuredPeakTflops":2250.0,"mfuConfiguredPct":29.688829406850836,"vramAllocatedGb":187.929870848,"vramAllocatedPct":98.13416147448893,"vramReservedGb":188.773040128,"vramReservedPct":98.57445183333652,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.296793088,"vramAllocatedPct":99.3701327878816,"vramReservedGb":190.54723072,"vramReservedPct":99.50090756523382,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 57.81 MiB is free. Including non-PyTorch memory, this process has 178.28 GiB memory in use. Of the allocated memory 177.23 GiB is allocated by PyTorch, and 238.84 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":9278.3504912331,"meanTps":9277.20191093826,"stepMs":882.9155578613281,"jitter":0.0005302361584794253,"achievedTflops":452.82635362154787,"nominalPeakTflops":2250.0,"mfuNominalPct":20.12561571651324,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.12561571651324,"vramAllocatedGb":187.929870848,"vramAllocatedPct":65.38314008856511,"vramReservedGb":188.653502464,"vramReservedPct":65.63490053041477,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":12341.166671003586,"meanTps":12339.244977024295,"stepMs":1327.5892333984375,"jitter":0.000889510023850614,"achievedTflops":602.3059280145418,"nominalPeakTflops":2250.0,"mfuNominalPct":26.769152356201857,"configuredPeakTflops":2250.0,"mfuConfiguredPct":26.769152356201857,"vramAllocatedGb":264.074863104,"vramAllocatedPct":91.87493020820769,"vramReservedGb":265.560260608,"vramReservedPct":92.39171847956126,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.196478976,"vramAllocatedPct":99.57131558333654,"vramReservedGb":286.426923008,"vramReservedPct":99.65148992900518,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 99.62 MiB is free. Including non-PyTorch memory, this process has 267.58 GiB memory in use. Of the allocated memory 266.54 GiB is allocated by PyTorch, and 219.77 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":16,"tokensPerStep":8192,"status":"complete","stable":true,"tps":9275.02792592885,"meanTps":9274.427335532017,"stepMs":883.2318420410156,"jitter":0.0006151856379338063,"achievedTflops":452.66419709029657,"nominalPeakTflops":2250.0,"mfuNominalPct":20.118408759568737,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.118408759568737,"vramAllocatedGb":187.929870848,"vramAllocatedPct":65.38314008856511,"vramReservedGb":188.653502464,"vramReservedPct":65.63490053041477,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":32,"tokensPerStep":16384,"status":"complete","stable":true,"tps":12336.765387277768,"meanTps":12331.38262294841,"stepMs":1328.0628662109375,"jitter":0.0016374093512279205,"achievedTflops":602.0911250425372,"nominalPeakTflops":2250.0,"mfuNominalPct":26.7596055574461,"configuredPeakTflops":2250.0,"mfuConfiguredPct":26.7596055574461,"vramAllocatedGb":264.074863104,"vramAllocatedPct":91.87493020820769,"vramReservedGb":265.560260608,"vramReservedPct":92.39171847956126,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":64,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.196478976,"vramAllocatedPct":99.57131558333654,"vramReservedGb":286.426923008,"vramReservedPct":99.65148992900518,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 99.62 MiB is free. Including non-PyTorch memory, this process has 267.58 GiB memory in use. Of the allocated memory 266.54 GiB is allocated by PyTorch, and 219.77 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":2,"tokensPerStep":1024,"status":"complete","stable":true,"tps":4392.232116674478,"meanTps":4391.476017120869,"stepMs":233.13886260986328,"jitter":0.0007905944493642363,"achievedTflops":214.36121167576502,"nominalPeakTflops":989.5,"mfuNominalPct":21.663588850506823,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.663588850506823,"vramAllocatedGb":128.408334336,"vramAllocatedPct":85.536693884429,"vramReservedGb":129.620770816,"vramReservedPct":86.34433467021091,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":6148.606482386454,"meanTps":6149.2741454349225,"stepMs":333.08360290527344,"jitter":0.0028667899584987233,"achievedTflops":300.0803920808602,"nominalPeakTflops":989.5,"mfuNominalPct":30.326467112770104,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.326467112770104,"vramAllocatedGb":130.87119616,"vramAllocatedPct":87.17728099280077,"vramReservedGb":131.992649728,"vramReservedPct":87.92431529589058,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.122546688,"vramAllocatedPct":99.33506024571074,"vramReservedGb":149.300445184,"vramReservedPct":99.45356384030636,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 500.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 33.44 MiB is free. Including non-PyTorch memory, this process has 139.77 GiB memory in use. Of the allocated memory 138.88 GiB is allocated by PyTorch, and 169.66 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":2,"tokensPerStep":1024,"status":"complete","stable":true,"tps":4360.540617424277,"meanTps":4360.7765635489195,"stepMs":234.83326721191406,"jitter":0.000612594748822176,"achievedTflops":212.81452015340574,"nominalPeakTflops":989.5,"mfuNominalPct":21.507278438949545,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.507278438949545,"vramAllocatedGb":128.408334336,"vramAllocatedPct":85.536693884429,"vramReservedGb":129.620770816,"vramReservedPct":86.34433467021091,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":4,"tokensPerStep":2048,"status":"complete","stable":true,"tps":6144.782912251546,"meanTps":6144.653928967706,"stepMs":333.2908630371094,"jitter":0.003938612826799069,"achievedTflops":299.8937842001121,"nominalPeakTflops":989.5,"mfuNominalPct":30.307608307237203,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.307608307237203,"vramAllocatedGb":130.87119616,"vramAllocatedPct":87.17728099280077,"vramReservedGb":131.992649728,"vramReservedPct":87.92431529589058,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":512,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.122546688,"vramAllocatedPct":99.33506024571074,"vramReservedGb":149.300445184,"vramReservedPct":99.45356384030636,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 500.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 33.44 MiB is free. Including non-PyTorch memory, this process has 139.77 GiB memory in use. Of the allocated memory 138.88 GiB is allocated by PyTorch, and 169.66 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":11383.227918097722,"meanTps":11375.097901803436,"stepMs":359.82763671875,"jitter":0.0011528633517973964,"achievedTflops":565.8669520538638,"nominalPeakTflops":2250.0,"mfuNominalPct":25.14964231350506,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.14964231350506,"vramAllocatedGb":149.863373312,"vramAllocatedPct":78.25640708073703,"vramReservedGb":150.330146816,"vramReservedPct":78.5001492075573,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":13418.384657873887,"meanTps":13422.267880350766,"stepMs":610.5056762695312,"jitter":0.0024864336754969367,"achievedTflops":667.0357900649249,"nominalPeakTflops":2250.0,"mfuNominalPct":29.646035113996657,"configuredPeakTflops":2250.0,"mfuConfiguredPct":29.646035113996657,"vramAllocatedGb":187.93714944,"vramAllocatedPct":98.13796224612472,"vramReservedGb":188.779331584,"vramReservedPct":98.57773713735035,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.30151168,"vramAllocatedPct":99.37259676589197,"vramReservedGb":190.549327872,"vramReservedPct":99.50200266657176,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 55.81 MiB is free. Including non-PyTorch memory, this process has 178.28 GiB memory in use. Of the allocated memory 177.23 GiB is allocated by PyTorch, and 236.34 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":4,"tokensPerStep":4096,"status":"complete","stable":true,"tps":11377.743537911683,"meanTps":11375.364323778824,"stepMs":360.00108337402344,"jitter":0.0012081731931772564,"achievedTflops":565.5943202905266,"nominalPeakTflops":2250.0,"mfuNominalPct":25.137525346245628,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.137525346245628,"vramAllocatedGb":149.863373312,"vramAllocatedPct":78.25640708073703,"vramReservedGb":150.330146816,"vramReservedPct":78.5001492075573,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":13426.197403354276,"meanTps":13424.664271314075,"stepMs":610.1504211425781,"jitter":0.0016503070086747087,"achievedTflops":667.4241662358993,"nominalPeakTflops":2250.0,"mfuNominalPct":29.66329627715108,"configuredPeakTflops":2250.0,"mfuConfiguredPct":29.66329627715108,"vramAllocatedGb":187.93714944,"vramAllocatedPct":98.13796224612472,"vramReservedGb":188.779331584,"vramReservedPct":98.57773713735035,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.30151168,"vramAllocatedPct":99.37259676589197,"vramReservedGb":190.549327872,"vramReservedPct":99.50200266657176,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 55.81 MiB is free. Including non-PyTorch memory, this process has 178.28 GiB memory in use. Of the allocated memory 177.23 GiB is allocated by PyTorch, and 236.34 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":9156.437365159507,"meanTps":9150.51040815059,"stepMs":894.6711120605469,"jitter":0.0008743403486006537,"achievedTflops":455.1718845282319,"nominalPeakTflops":2250.0,"mfuNominalPct":20.229861534588085,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.229861534588085,"vramAllocatedGb":187.93714944,"vramAllocatedPct":65.38567240127429,"vramReservedGb":188.678668288,"vramReservedPct":65.64365603367038,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":12144.184547679602,"meanTps":12142.036238838115,"stepMs":1349.1231079101562,"jitter":0.0009543921419226979,"achievedTflops":603.6945534798253,"nominalPeakTflops":2250.0,"mfuNominalPct":26.830869043547793,"configuredPeakTflops":2250.0,"mfuConfiguredPct":26.830869043547793,"vramAllocatedGb":264.084701696,"vramAllocatedPct":91.87835317676563,"vramReservedGb":265.585426432,"vramReservedPct":92.40047398281686,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.201201664,"vramAllocatedPct":99.57295866524633,"vramReservedGb":286.431117312,"vramReservedPct":99.65294917954779,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 95.62 MiB is free. Including non-PyTorch memory, this process has 267.58 GiB memory in use. Of the allocated memory 266.55 GiB is allocated by PyTorch, and 219.26 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":8,"tokensPerStep":8192,"status":"complete","stable":true,"tps":9155.843977329932,"meanTps":9156.208402037184,"stepMs":894.7290954589844,"jitter":0.00047094707458090513,"achievedTflops":455.14238687036857,"nominalPeakTflops":2250.0,"mfuNominalPct":20.228550527571937,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.228550527571937,"vramAllocatedGb":187.93714944,"vramAllocatedPct":65.38567240127429,"vramReservedGb":188.678668288,"vramReservedPct":65.64365603367038,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":16,"tokensPerStep":16384,"status":"complete","stable":true,"tps":12144.482884829498,"meanTps":12144.290365035391,"stepMs":1349.0899658203125,"jitter":0.0010367841549004252,"achievedTflops":603.7093839949404,"nominalPeakTflops":2250.0,"mfuNominalPct":26.831528177552904,"configuredPeakTflops":2250.0,"mfuConfiguredPct":26.831528177552904,"vramAllocatedGb":264.084701696,"vramAllocatedPct":91.87835317676563,"vramReservedGb":265.585426432,"vramReservedPct":92.40047398281686,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":32,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.201201664,"vramAllocatedPct":99.57295866524633,"vramReservedGb":286.431117312,"vramReservedPct":99.65294917954779,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 95.62 MiB is free. Including non-PyTorch memory, this process has 267.58 GiB memory in use. Of the allocated memory 266.55 GiB is allocated by PyTorch, and 219.26 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":1,"tokensPerStep":1024,"status":"complete","stable":true,"tps":4355.140565985114,"meanTps":4356.3393489873915,"stepMs":235.12444305419922,"jitter":0.000718154549493772,"achievedTflops":216.49659793967933,"nominalPeakTflops":989.5,"mfuNominalPct":21.879393424929695,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.879393424929695,"vramAllocatedGb":128.413052928,"vramAllocatedPct":85.53983708196019,"vramReservedGb":129.62496512,"vramReservedPct":86.34712862357198,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":6077.328270346886,"meanTps":6076.734079095815,"stepMs":336.9901885986328,"jitter":0.0017826788593827236,"achievedTflops":302.1075612045431,"nominalPeakTflops":989.5,"mfuNominalPct":30.531335139418204,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.531335139418204,"vramAllocatedGb":130.876554752,"vramAllocatedPct":87.18085051378183,"vramReservedGb":131.998941184,"vramReservedPct":87.92850622593217,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.12777728,"vramAllocatedPct":99.33854450200182,"vramReservedGb":149.304639488,"vramReservedPct":99.45635779366742,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 500.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 29.44 MiB is free. Including non-PyTorch memory, this process has 139.77 GiB memory in use. Of the allocated memory 138.89 GiB is allocated by PyTorch, and 168.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":1,"tokensPerStep":1024,"status":"complete","stable":true,"tps":4325.981050128888,"meanTps":4324.8708508611135,"stepMs":236.70931243896484,"jitter":0.0008280320839678691,"achievedTflops":215.04706126347028,"nominalPeakTflops":989.5,"mfuNominalPct":21.73290159307431,"configuredPeakTflops":989.5,"mfuConfiguredPct":21.73290159307431,"vramAllocatedGb":128.413052928,"vramAllocatedPct":85.53983708196019,"vramReservedGb":129.62496512,"vramReservedPct":86.34712862357198,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":2,"tokensPerStep":2048,"status":"complete","stable":true,"tps":6064.255174008259,"meanTps":6067.300557364358,"stepMs":337.71665954589844,"jitter":0.0029178938000006234,"achievedTflops":301.4576898998244,"nominalPeakTflops":989.5,"mfuNominalPct":30.46565840321621,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.46565840321621,"vramAllocatedGb":130.876554752,"vramAllocatedPct":87.18085051378183,"vramReservedGb":131.998941184,"vramReservedPct":87.92850622593217,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":1024,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.12777728,"vramAllocatedPct":99.33854450200182,"vramReservedGb":149.304639488,"vramReservedPct":99.45635779366742,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 500.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 29.44 MiB is free. Including non-PyTorch memory, this process has 139.77 GiB memory in use. Of the allocated memory 138.89 GiB is allocated by PyTorch, and 168.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.274397184,"vramAllocatedPct":99.17640621191975,"vramReservedGb":84.416659456,"vramReservedPct":99.34382432878505,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 64.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 6.75 MiB is free. Process 703773 has 79.12 GiB memory in use. Of the allocated memory 78.49 GiB is allocated by PyTorch, and 135.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a100_pcie_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA A100 80GB PCIe","gpuUuid":"GPU-2e75422d-0492-4288-c66f-44c400810994","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":312.0,"mfuNominalPct":null,"configuredPeakTflops":312.0,"mfuConfiguredPct":null,"vramAllocatedGb":84.274397184,"vramAllocatedPct":99.17640621191975,"vramReservedGb":84.416659456,"vramReservedPct":99.34382432878505,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 64.00 MiB. GPU 0 has a total capacity of 79.14 GiB of which 6.75 MiB is free. Process 736416 has 79.12 GiB memory in use. Of the allocated memory 78.49 GiB is allocated by PyTorch, and 135.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":7965.346417723321,"meanTps":7947.108945351409,"stepMs":257.1137390136719,"jitter":0.004655747021727805,"achievedTflops":410.39477771946343,"nominalPeakTflops":2250.0,"mfuNominalPct":18.23976789864282,"configuredPeakTflops":2250.0,"mfuConfiguredPct":18.23976789864282,"vramAllocatedGb":130.836226048,"vramAllocatedPct":68.3207159977879,"vramReservedGb":131.898277888,"vramReservedPct":68.87530354840212,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":10896.099443730987,"meanTps":10886.476883038435,"stepMs":375.9143371582031,"jitter":0.002683504867404597,"achievedTflops":561.3945803122099,"nominalPeakTflops":2250.0,"mfuNominalPct":24.950870236098215,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.950870236098215,"vramAllocatedGb":149.873450496,"vramAllocatedPct":78.26166923516412,"vramReservedGb":150.430810112,"vramReservedPct":78.55271407177841,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":12870.69902014847,"meanTps":12872.948356510187,"stepMs":636.4844665527344,"jitter":0.0014341672307242267,"achievedTflops":663.1309407604753,"nominalPeakTflops":2250.0,"mfuNominalPct":29.47248625602112,"configuredPeakTflops":2250.0,"mfuConfiguredPct":29.47248625602112,"vramAllocatedGb":187.947866624,"vramAllocatedPct":98.14355859895815,"vramReservedGb":188.756262912,"vramReservedPct":98.565691022633,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.310948864,"vramAllocatedPct":99.3775247219127,"vramReservedGb":190.446567424,"vramReservedPct":99.44834270101269,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 153.81 MiB is free. Including non-PyTorch memory, this process has 178.19 GiB memory in use. Of the allocated memory 177.24 GiB is allocated by PyTorch, and 129.34 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":7781.424875947019,"meanTps":7767.082828168305,"stepMs":263.1908721923828,"jitter":0.013292231343091914,"achievedTflops":400.9186750747425,"nominalPeakTflops":2250.0,"mfuNominalPct":17.818607781099665,"configuredPeakTflops":2250.0,"mfuConfiguredPct":17.818607781099665,"vramAllocatedGb":130.836226048,"vramAllocatedPct":68.3207159977879,"vramReservedGb":131.898277888,"vramReservedPct":68.87530354840212,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":10860.419880229114,"meanTps":10859.537386718208,"stepMs":377.1493225097656,"jitter":0.0019960660129212217,"achievedTflops":559.5562790300587,"nominalPeakTflops":2250.0,"mfuNominalPct":24.8691679568915,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.8691679568915,"vramAllocatedGb":149.873450496,"vramAllocatedPct":78.26166923516412,"vramReservedGb":150.430810112,"vramReservedPct":78.55271407177841,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":12837.12027778078,"meanTps":12826.564476851443,"stepMs":638.1493530273438,"jitter":0.0019098574531399397,"achievedTflops":661.4008790924196,"nominalPeakTflops":2250.0,"mfuNominalPct":29.39559462632976,"configuredPeakTflops":2250.0,"mfuConfiguredPct":29.39559462632976,"vramAllocatedGb":187.947866624,"vramAllocatedPct":98.14355859895815,"vramReservedGb":188.756262912,"vramReservedPct":98.565691022633,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.310948864,"vramAllocatedPct":99.3775247219127,"vramReservedGb":190.446567424,"vramReservedPct":99.44834270101269,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 153.81 MiB is free. Including non-PyTorch memory, this process has 178.19 GiB memory in use. Of the allocated memory 177.24 GiB is allocated by PyTorch, and 129.34 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":3689.6037894428264,"meanTps":3689.1967101329674,"stepMs":555.0731506347656,"jitter":0.0006187608560223616,"achievedTflops":190.09771171685844,"nominalPeakTflops":2250.0,"mfuNominalPct":8.448787187415931,"configuredPeakTflops":2250.0,"mfuConfiguredPct":8.448787187415931,"vramAllocatedGb":130.836226048,"vramAllocatedPct":45.519550765160304,"vramReservedGb":131.898277888,"vramReservedPct":45.8890518131986,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":6051.7918061806695,"meanTps":6050.666338994081,"stepMs":676.8243408203125,"jitter":0.0007997872864229616,"achievedTflops":311.803608136338,"nominalPeakTflops":2250.0,"mfuNominalPct":13.857938139392802,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.857938139392802,"vramAllocatedGb":149.873450496,"vramAllocatedPct":52.142837991211664,"vramReservedGb":150.430810112,"vramReservedPct":52.33675033568463,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":8932.613960224171,"meanTps":8931.442782886112,"stepMs":917.0887756347656,"jitter":0.0003568462587057773,"achievedTflops":460.2308460185932,"nominalPeakTflops":2250.0,"mfuNominalPct":20.454704267493028,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.454704267493028,"vramAllocatedGb":187.947866624,"vramAllocatedPct":65.38940104291952,"vramReservedGb":188.756262912,"vramReservedPct":65.67065216870851,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":11731.761519896323,"meanTps":11722.269135916278,"stepMs":1396.5507202148438,"jitter":0.0008657376548587923,"achievedTflops":604.450002388188,"nominalPeakTflops":2250.0,"mfuNominalPct":26.864444550586132,"configuredPeakTflops":2250.0,"mfuConfiguredPct":26.864444550586132,"vramAllocatedGb":264.09669888,"vramAllocatedPct":91.88252714633524,"vramReservedGb":265.545580544,"vramReservedPct":92.38661110266216,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.344864768,"vramAllocatedPct":99.62294084642916,"vramReservedGb":286.496129024,"vramReservedPct":99.67556756295811,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 53.62 MiB is free. Including non-PyTorch memory, this process has 267.62 GiB memory in use. Of the allocated memory 266.68 GiB is allocated by PyTorch, and 124.26 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":3687.3976724329914,"meanTps":3684.20488039417,"stepMs":555.4052429199219,"jitter":0.0010221249363788132,"achievedTflops":189.98404699314221,"nominalPeakTflops":2250.0,"mfuNominalPct":8.443735421917433,"configuredPeakTflops":2250.0,"mfuConfiguredPct":8.443735421917433,"vramAllocatedGb":130.836226048,"vramAllocatedPct":45.519550765160304,"vramReservedGb":131.898277888,"vramReservedPct":45.8890518131986,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":2,"tokensPerStep":4096,"status":"complete","stable":true,"tps":6049.398579002777,"meanTps":6049.678009429508,"stepMs":677.0921020507812,"jitter":0.000498165859089986,"achievedTflops":311.6803030238927,"nominalPeakTflops":2250.0,"mfuNominalPct":13.852457912173008,"configuredPeakTflops":2250.0,"mfuConfiguredPct":13.852457912173008,"vramAllocatedGb":149.873450496,"vramAllocatedPct":52.142837991211664,"vramReservedGb":150.430810112,"vramReservedPct":52.33675033568463,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":4,"tokensPerStep":8192,"status":"complete","stable":true,"tps":8933.946716645785,"meanTps":8931.869181185015,"stepMs":916.9519653320312,"jitter":0.0007554446777184915,"achievedTflops":460.29951299761944,"nominalPeakTflops":2250.0,"mfuNominalPct":20.45775613322753,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.45775613322753,"vramAllocatedGb":187.947866624,"vramAllocatedPct":65.38940104291952,"vramReservedGb":188.756262912,"vramReservedPct":65.67065216870851,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":8,"tokensPerStep":16384,"status":"complete","stable":true,"tps":11717.039615648056,"meanTps":11692.595077981421,"stepMs":1398.305419921875,"jitter":0.003229774355434059,"achievedTflops":603.6914926756498,"nominalPeakTflops":2250.0,"mfuNominalPct":26.830733007806657,"configuredPeakTflops":2250.0,"mfuConfiguredPct":26.830733007806657,"vramAllocatedGb":264.09669888,"vramAllocatedPct":91.88252714633524,"vramReservedGb":265.545580544,"vramReservedPct":92.38661110266216,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":16,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.344864768,"vramAllocatedPct":99.62294084642916,"vramReservedGb":286.496129024,"vramReservedPct":99.67556756295811,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 53.62 MiB is free. Including non-PyTorch memory, this process has 267.62 GiB memory in use. Of the allocated memory 266.68 GiB is allocated by PyTorch, and 124.26 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":null,"context":2048,"batch":null,"tokensPerStep":null,"status":"oom_during_model_build","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":null,"vramAllocatedPct":null,"vramReservedGb":null,"vramReservedPct":null,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 449.69 MiB is free. Including non-PyTorch memory, this process has 23.08 GiB memory in use. Of the allocated memory 22.60 GiB is allocated by PyTorch, and 20.86 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx4090_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 4090","gpuUuid":"GPU-5447a6b8-ccd6-9d49-4872-492ff5116ee3","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":null,"context":2048,"batch":null,"tokensPerStep":null,"status":"oom_during_model_build","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":165.2,"mfuNominalPct":null,"configuredPeakTflops":205.51666666666665,"mfuConfiguredPct":null,"vramAllocatedGb":null,"vramAllocatedPct":null,"vramReservedGb":null,"vramReservedPct":null,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 23.53 GiB of which 449.69 MiB is free. Including non-PyTorch memory, this process has 23.08 GiB memory in use. Of the allocated memory 22.60 GiB is allocated by PyTorch, and 20.86 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed11.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.99155456,"vramAllocatedPct":97.98833896030334,"vramReservedGb":33.013366784,"vramReservedPct":98.05312352796908,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 32.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 13.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.73 GiB is allocated by PyTorch, and 20.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx5090_training_seed22.json","sourceKind":"baseline_reused","gpu":"NVIDIA GeForce RTX 5090","gpuUuid":"GPU-c3733010-4951-8bd3-2692-36d9198c8d52","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":209.5,"mfuNominalPct":null,"configuredPeakTflops":268.946821769838,"mfuConfiguredPct":null,"vramAllocatedGb":32.99155456,"vramAllocatedPct":97.98833896030334,"vramReservedGb":33.013366784,"vramReservedPct":98.05312352796908,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 32.00 MiB. GPU 0 has a total capacity of 31.36 GiB of which 13.88 MiB is free. Including non-PyTorch memory, this process has 31.33 GiB memory in use. Of the allocated memory 30.73 GiB is allocated by PyTorch, and 20.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.854670848,"vramAllocatedPct":98.63225483594704,"vramReservedGb":83.997229056,"vramReservedPct":98.79993586502053,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 220.19 MiB is free. Including non-PyTorch memory, this process has 78.96 GiB memory in use. Of the allocated memory 78.10 GiB is allocated by PyTorch, and 135.95 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h100_sxm_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H100 80GB HBM3","gpuUuid":"GPU-8f62317d-f779-de5a-419c-ecd52f14c564","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":83.854670848,"vramAllocatedPct":98.63225483594704,"vramReservedGb":83.997229056,"vramReservedPct":98.79993586502053,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 79.18 GiB of which 220.19 MiB is free. Including non-PyTorch memory, this process has 78.96 GiB memory in use. Of the allocated memory 78.10 GiB is allocated by PyTorch, and 135.95 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":5945.414571225452,"meanTps":5943.629323423078,"stepMs":344.46714782714844,"jitter":0.0026489121173383,"achievedTflops":306.32278415149295,"nominalPeakTflops":989.5,"mfuNominalPct":30.95733038418322,"configuredPeakTflops":989.5,"mfuConfiguredPct":30.95733038418322,"vramAllocatedGb":130.886295552,"vramAllocatedPct":87.18733915668884,"vramReservedGb":131.940220928,"vramReservedPct":87.88939087887732,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.137470464,"vramAllocatedPct":99.34500142644416,"vramReservedGb":149.277376512,"vramReservedPct":99.43819709682053,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 500.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 53.44 MiB is free. Including non-PyTorch memory, this process has 139.75 GiB memory in use. Of the allocated memory 138.90 GiB is allocated by PyTorch, and 133.42 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":2048,"status":"complete","stable":true,"tps":5960.191622827077,"meanTps":5963.508371546546,"stepMs":343.6131134033203,"jitter":0.0049777215349738205,"achievedTflops":307.0841351950464,"nominalPeakTflops":989.5,"mfuNominalPct":31.034273390100694,"configuredPeakTflops":989.5,"mfuConfiguredPct":31.034273390100694,"vramAllocatedGb":130.886295552,"vramAllocatedPct":87.18733915668884,"vramReservedGb":131.940220928,"vramReservedPct":87.88939087887732,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"h200_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.137470464,"vramAllocatedPct":99.34500142644416,"vramReservedGb":149.277376512,"vramReservedPct":99.43819709682053,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 500.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 53.44 MiB is free. Including non-PyTorch memory, this process has 139.75 GiB memory in use. Of the allocated memory 138.90 GiB is allocated by PyTorch, and 133.42 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":50.206611456,"vramAllocatedPct":98.68264125314465,"vramReservedGb":50.224693248,"vramReservedPct":98.71818157226603,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 96.50 MiB is free. Process 2308332 has 47.28 GiB memory in use. Of the allocated memory 46.76 GiB is allocated by PyTorch, and 17.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtx6000_ada_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX 6000 Ada Generation","gpuUuid":"GPU-5b377adf-91ee-db42-9c14-0b8ba36df6b5","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":364.2,"mfuNominalPct":null,"configuredPeakTflops":451.43353293413173,"mfuConfiguredPct":null,"vramAllocatedGb":50.206611456,"vramAllocatedPct":98.68264125314465,"vramReservedGb":50.224693248,"vramReservedPct":98.71818157226603,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 224.00 MiB. GPU 0 has a total capacity of 47.38 GiB of which 96.50 MiB is free. Process 2312978 has 47.28 GiB memory in use. Of the allocated memory 46.76 GiB is allocated by PyTorch, and 17.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.634430464,"vramAllocatedPct":99.21718492364353,"vramReservedGb":50.665095168,"vramReservedPct":99.27727181668284,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 14.25 MiB is free. Process 1348507 has 47.51 GiB memory in use. Of the allocated memory 47.16 GiB is allocated by PyTorch, and 29.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"a6000_training_seed22_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX A6000","gpuUuid":"GPU-18d70bd6-a015-e960-7fda-6a87d5e18b1f","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":154.8,"mfuNominalPct":null,"configuredPeakTflops":180.6,"mfuConfiguredPct":null,"vramAllocatedGb":50.634430464,"vramAllocatedPct":99.21718492364353,"vramReservedGb":50.665095168,"vramReservedPct":99.27727181668284,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 56.00 MiB. GPU 0 has a total capacity of 47.53 GiB of which 14.25 MiB is free. Process 1380474 has 47.51 GiB memory in use. Of the allocated memory 47.16 GiB is allocated by PyTorch, and 29.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed11_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.783839744,"vramAllocatedPct":98.83292672826268,"vramReservedGb":100.921245696,"vramReservedPct":98.96767285840158,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 333.75 MiB is free. Including non-PyTorch memory, this process has 94.64 GiB memory in use. Of the allocated memory 93.86 GiB is allocated by PyTorch, and 131.04 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"rtxpro6000_server_training_seed22_merged_v2.json","sourceKind":"baseline_reused","gpu":"NVIDIA RTX PRO 6000 Blackwell Server Edition","gpuUuid":"GPU-0adc4b73-c849-be4e-7fc3-492d46abdd3e","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":2048,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":468.0,"mfuNominalPct":null,"configuredPeakTflops":468.0,"mfuConfiguredPct":null,"vramAllocatedGb":100.783839744,"vramAllocatedPct":98.83292672826268,"vramReservedGb":100.921245696,"vramReservedPct":98.96767285840158,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 335.75 MiB is free. Including non-PyTorch memory, this process has 94.63 GiB memory in use. Of the allocated memory 93.86 GiB is allocated by PyTorch, and 131.04 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":10193.966470229614,"meanTps":10195.282541189312,"stepMs":401.8063049316406,"jitter":0.0009785366359751334,"achievedTflops":562.1606125759449,"nominalPeakTflops":2250.0,"mfuNominalPct":24.984916114486442,"configuredPeakTflops":2250.0,"mfuConfiguredPct":24.984916114486442,"vramAllocatedGb":149.892612096,"vramAllocatedPct":78.27167513545035,"vramReservedGb":150.084780032,"vramReservedPct":78.3720223510183,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":11977.229671282303,"meanTps":11977.439162846536,"stepMs":683.9645080566406,"jitter":0.001094743185939019,"achievedTflops":660.5011688663307,"nominalPeakTflops":2250.0,"mfuNominalPct":29.355607505170255,"configuredPeakTflops":2250.0,"mfuConfiguredPct":29.355607505170255,"vramAllocatedGb":187.967380992,"vramAllocatedPct":98.15374870940596,"vramReservedGb":188.804497408,"vramReservedPct":98.59087835340563,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.329823232,"vramAllocatedPct":99.38738063395417,"vramReservedGb":190.582882304,"vramReservedPct":99.5195242879788,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 23.81 MiB is free. Including non-PyTorch memory, this process has 178.31 GiB memory in use. Of the allocated memory 177.26 GiB is allocated by PyTorch, and 241.34 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":1,"tokensPerStep":4096,"status":"complete","stable":true,"tps":10202.125372348344,"meanTps":10200.135797401172,"stepMs":401.48497009277344,"jitter":0.0006341912894402681,"achievedTflops":562.6105467037848,"nominalPeakTflops":2250.0,"mfuNominalPct":25.004913186834884,"configuredPeakTflops":2250.0,"mfuConfiguredPct":25.004913186834884,"vramAllocatedGb":149.892612096,"vramAllocatedPct":78.27167513545035,"vramReservedGb":150.084780032,"vramReservedPct":78.3720223510183,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":11976.778647992262,"meanTps":11972.671143910728,"stepMs":683.9902648925781,"jitter":0.001876584684606826,"achievedTflops":660.4762965529131,"nominalPeakTflops":2250.0,"mfuNominalPct":29.35450206901836,"configuredPeakTflops":2250.0,"mfuConfiguredPct":29.35450206901836,"vramAllocatedGb":187.967380992,"vramAllocatedPct":98.15374870940596,"vramReservedGb":188.804497408,"vramReservedPct":98.59087835340563,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.329823232,"vramAllocatedPct":99.38738063395417,"vramReservedGb":190.582882304,"vramReservedPct":99.5195242879788,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 23.81 MiB is free. Including non-PyTorch memory, this process has 178.31 GiB memory in use. Of the allocated memory 177.26 GiB is allocated by PyTorch, and 241.34 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":8456.208782805417,"meanTps":8453.087224794315,"stepMs":968.7556457519531,"jitter":0.0016317659061021725,"achievedTflops":466.329521810258,"nominalPeakTflops":2250.0,"mfuNominalPct":20.72575652490035,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.72575652490035,"vramAllocatedGb":187.967380992,"vramAllocatedPct":65.39619033432342,"vramReservedGb":188.708028416,"vramReservedPct":65.65387078746859,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":11047.754980319567,"meanTps":11048.343734526543,"stepMs":1483.0162353515625,"jitter":0.0004332279819627654,"achievedTflops":609.2439802958762,"nominalPeakTflops":2250.0,"mfuNominalPct":27.077510235372277,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.077510235372277,"vramAllocatedGb":264.116853248,"vramAllocatedPct":91.88953910170133,"vramReservedGb":265.612689408,"vramReservedPct":92.40995911134378,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.229537792,"vramAllocatedPct":99.58281715670503,"vramReservedGb":286.458380288,"vramReservedPct":99.6624343080747,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 69.62 MiB is free. Including non-PyTorch memory, this process has 267.61 GiB memory in use. Of the allocated memory 266.57 GiB is allocated by PyTorch, and 218.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":2,"tokensPerStep":8192,"status":"complete","stable":true,"tps":8460.237926315325,"meanTps":8459.572738523155,"stepMs":968.2942810058594,"jitter":0.00039404769765212446,"achievedTflops":466.55171459363635,"nominalPeakTflops":2250.0,"mfuNominalPct":20.735631759717172,"configuredPeakTflops":2250.0,"mfuConfiguredPct":20.735631759717172,"vramAllocatedGb":187.967380992,"vramAllocatedPct":65.39619033432342,"vramReservedGb":188.708028416,"vramReservedPct":65.65387078746859,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":4,"tokensPerStep":16384,"status":"complete","stable":true,"tps":11039.032878155373,"meanTps":11038.229282780556,"stepMs":1484.18798828125,"jitter":0.0005793965078495086,"achievedTflops":608.762987709733,"nominalPeakTflops":2250.0,"mfuNominalPct":27.056132787099244,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.056132787099244,"vramAllocatedGb":264.116853248,"vramAllocatedPct":91.88953910170133,"vramReservedGb":265.612689408,"vramReservedPct":92.40995911134378,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":8,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.229537792,"vramAllocatedPct":99.58281715670503,"vramReservedGb":286.458380288,"vramReservedPct":99.6624343080747,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 69.62 MiB is free. Including non-PyTorch memory, this process has 267.61 GiB memory in use. Of the allocated memory 266.57 GiB is allocated by PyTorch, and 218.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.894360064,"vramAllocatedPct":99.18305820077698,"vramReservedGb":149.078147072,"vramReservedPct":99.30548431217015,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 500.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 305.44 MiB is free. Including non-PyTorch memory, this process has 139.50 GiB memory in use. Of the allocated memory 138.67 GiB is allocated by PyTorch, and 115.27 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":4096,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":148.894360064,"vramAllocatedPct":99.18305820077698,"vramReservedGb":149.078147072,"vramReservedPct":99.30548431217015,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 500.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 305.44 MiB is free. Including non-PyTorch memory, this process has 139.50 GiB memory in use. Of the allocated memory 138.67 GiB is allocated by PyTorch, and 115.27 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10428.774082371616,"meanTps":10427.585599723561,"stepMs":785.5189819335938,"jitter":0.000909320507480103,"achievedTflops":650.6946320714248,"nominalPeakTflops":2250.0,"mfuNominalPct":28.91976142539666,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.91976142539666,"vramAllocatedGb":188.005384192,"vramAllocatedPct":98.17359341077525,"vramReservedGb":188.219392,"vramReservedPct":98.28534508012035,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.099070976,"vramAllocatedPct":99.26688526486394,"vramReservedGb":190.53674496,"vramReservedPct":99.49543205854411,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 67.81 MiB is free. Including non-PyTorch memory, this process has 178.27 GiB memory in use. Of the allocated memory 177.04 GiB is allocated by PyTorch, and 417.40 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":10416.067556737371,"meanTps":10411.444168934446,"stepMs":786.4772338867188,"jitter":0.0017804119204606241,"achievedTflops":649.9018190372967,"nominalPeakTflops":2250.0,"mfuNominalPct":28.88452529054652,"configuredPeakTflops":2250.0,"mfuConfiguredPct":28.88452529054652,"vramAllocatedGb":188.005384192,"vramAllocatedPct":98.17359341077525,"vramReservedGb":188.219392,"vramReservedPct":98.28534508012035,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":8192,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.099070976,"vramAllocatedPct":99.26688526486394,"vramReservedGb":190.53674496,"vramReservedPct":99.49543205854411,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 67.81 MiB is free. Including non-PyTorch memory, this process has 178.27 GiB memory in use. Of the allocated memory 177.04 GiB is allocated by PyTorch, and 417.40 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":7671.964519735866,"meanTps":7672.001005939711,"stepMs":1067.7838745117188,"jitter":0.00017389938855888492,"achievedTflops":478.6858063090092,"nominalPeakTflops":2250.0,"mfuNominalPct":21.274924724844855,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.274924724844855,"vramAllocatedGb":188.005384192,"vramAllocatedPct":65.4094121203982,"vramReservedGb":188.219392,"vramReservedPct":65.48386809925549,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":9772.777119751749,"meanTps":9770.85447819975,"stepMs":1676.4937744140625,"jitter":0.0004770954176898998,"achievedTflops":609.7642505270167,"nominalPeakTflops":2250.0,"mfuNominalPct":27.100633356756294,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.100633356756294,"vramAllocatedGb":264.155241984,"vramAllocatedPct":91.90289502054694,"vramReservedGb":265.6567296,"vramReservedPct":92.4252812420411,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.267319296,"vramAllocatedPct":99.59596181198332,"vramReservedGb":286.50242048,"vramReservedPct":99.67775643877201,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 27.62 MiB is free. Including non-PyTorch memory, this process has 267.65 GiB memory in use. Of the allocated memory 266.61 GiB is allocated by PyTorch, and 224.21 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":8192,"batch":1,"tokensPerStep":8192,"status":"complete","stable":true,"tps":7674.5141264740405,"meanTps":7673.905348213667,"stepMs":1067.4291381835938,"jitter":0.0004737196504337443,"achievedTflops":478.84488688792675,"nominalPeakTflops":2250.0,"mfuNominalPct":21.281994972796745,"configuredPeakTflops":2250.0,"mfuConfiguredPct":21.281994972796745,"vramAllocatedGb":188.005384192,"vramAllocatedPct":65.4094121203982,"vramReservedGb":188.219392,"vramReservedPct":65.48386809925549,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":8192,"batch":2,"tokensPerStep":16384,"status":"complete","stable":true,"tps":9774.428273292748,"meanTps":9771.678877874265,"stepMs":1676.2105712890625,"jitter":0.0006801607353360884,"achievedTflops":609.8672728705219,"nominalPeakTflops":2250.0,"mfuNominalPct":27.105212127578753,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.105212127578753,"vramAllocatedGb":264.155241984,"vramAllocatedPct":91.90289502054694,"vramReservedGb":265.6567296,"vramReservedPct":92.4252812420411,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-589e52a7-94b4-3cdf-33cc-048fbf864c10","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":8192,"batch":4,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.267319296,"vramAllocatedPct":99.59596181198332,"vramReservedGb":286.50242048,"vramReservedPct":99.67775643877201,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 27.62 MiB is free. Including non-PyTorch memory, this process has 267.65 GiB memory in use. Of the allocated memory 266.61 GiB is allocated by PyTorch, and 224.21 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.022614528,"vramAllocatedPct":99.26849239695441,"vramReservedGb":149.294153728,"vramReservedPct":99.44937291026477,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 79.44 MiB is free. Including non-PyTorch memory, this process has 139.72 GiB memory in use. Of the allocated memory 138.79 GiB is allocated by PyTorch, and 218.96 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"h200_context_frontier_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA H200","gpuUuid":"GPU-b23aad41-8254-c13e-e269-0db430f83167","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":8192,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":989.5,"mfuNominalPct":null,"configuredPeakTflops":989.5,"mfuConfiguredPct":null,"vramAllocatedGb":149.022614528,"vramAllocatedPct":99.26849239695441,"vramReservedGb":149.294153728,"vramReservedPct":99.44937291026477,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 448.00 MiB. GPU 0 has a total capacity of 139.81 GiB of which 79.44 MiB is free. Including non-PyTorch memory, this process has 139.72 GiB memory in use. Of the allocated memory 138.79 GiB is allocated by PyTorch, and 218.96 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.507390976,"vramAllocatedPct":99.48010384811766,"vramReservedGb":190.576590848,"vramReservedPct":99.51623898396498,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 31.81 MiB is free. Including non-PyTorch memory, this process has 178.31 GiB memory in use. Of the allocated memory 177.42 GiB is allocated by PyTorch, and 63.99 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b200_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B200","gpuUuid":"GPU-c4a7ec32-9030-147d-7e07-b6668d499a32","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":16384,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":190.507390976,"vramAllocatedPct":99.48010384811766,"vramReservedGb":190.576590848,"vramReservedPct":99.51623898396498,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacity of 178.35 GiB of which 31.81 MiB is free. Including non-PyTorch memory, this process has 178.31 GiB memory in use. Of the allocated memory 177.42 GiB is allocated by PyTorch, and 63.99 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":7962.518222533193,"meanTps":7962.676756232735,"stepMs":2057.6405029296875,"jitter":0.0003257003791663253,"achievedTflops":612.2354607983063,"nominalPeakTflops":2250.0,"mfuNominalPct":27.210464924369173,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.210464924369173,"vramAllocatedGb":264.230928384,"vramAllocatedPct":91.92922726071541,"vramReservedGb":264.98564096,"vramReservedPct":92.19180115522481,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.38644992,"vramAllocatedPct":99.28949639724722,"vramReservedGb":286.39756288,"vramReservedPct":99.64127517520697,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 127.62 MiB is free. Including non-PyTorch memory, this process has 267.55 GiB memory in use. Of the allocated memory 265.79 GiB is allocated by PyTorch, and 964.27 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":16384,"batch":1,"tokensPerStep":16384,"status":"complete","stable":true,"tps":7962.034063038642,"meanTps":7961.64205507566,"stepMs":2057.765625,"jitter":0.0004507461374404884,"achievedTflops":612.1982339307549,"nominalPeakTflops":2250.0,"mfuNominalPct":27.208810396922438,"configuredPeakTflops":2250.0,"mfuConfiguredPct":27.208810396922438,"vramAllocatedGb":264.230928384,"vramAllocatedPct":91.92922726071541,"vramReservedGb":264.98564096,"vramReservedPct":92.19180115522481,"warmupSteps":11,"measuredSteps":20,"sdpaBackend":"FLASH_ATTENTION","error":null},{"sourceFile":"b300_context_extension_16384_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":16384,"batch":2,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":285.38644992,"vramAllocatedPct":99.28949639724722,"vramReservedGb":286.39756288,"vramReservedPct":99.64127517520697,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 256.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 127.62 MiB is free. Including non-PyTorch memory, this process has 267.55 GiB memory in use. Of the allocated memory 265.79 GiB is allocated by PyTorch, and 964.27 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed11_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":11,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.056972288,"vramAllocatedPct":99.52277947098976,"vramReservedGb":286.18784768,"vramReservedPct":99.56831264807688,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 387.62 MiB is free. Including non-PyTorch memory, this process has 267.29 GiB memory in use. Of the allocated memory 266.41 GiB is allocated by PyTorch, and 64.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"},{"sourceFile":"b300_context_extension_32768_seed22_v2.json","sourceKind":"context_frontier","gpu":"NVIDIA B300 SXM6 AC","gpuUuid":"GPU-5cb378b4-bafb-0312-4b78-09892a22373d","seed":22,"modelKey":"8b","modelLabel":"8B","parameters":7983108096,"context":32768,"batch":1,"tokensPerStep":null,"status":"oom","stable":null,"tps":null,"meanTps":null,"stepMs":null,"jitter":null,"achievedTflops":null,"nominalPeakTflops":2250.0,"mfuNominalPct":null,"configuredPeakTflops":2250.0,"mfuConfiguredPct":null,"vramAllocatedGb":286.056972288,"vramAllocatedPct":99.52277947098976,"vramReservedGb":286.18784768,"vramReservedPct":99.56831264807688,"warmupSteps":null,"measuredSteps":null,"sdpaBackend":null,"error":"CUDA out of memory. Tried to allocate 896.00 MiB. GPU 0 has a total capacity of 267.69 GiB of which 387.62 MiB is free. Including non-PyTorch memory, this process has 267.29 GiB memory in use. Of the allocated memory 266.41 GiB is allocated by PyTorch, and 64.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"}],"interpolation":{"schemaVersion":1,"target":"Median steady-state tokens/second at the measured batch maximizing TPS","method":"local_power_law_interpolation","formula":{"exponent":"alpha = ln(T_upper / T_lower) / ln(P_upper / P_lower)","prediction":"TPS(P) = T_lower * (P / P_lower) ** alpha","equivalent":"Linear interpolation in log(parameter)-log(TPS) space"},"policy":{"interpolationOnly":true,"exactMeasuredContextsOnly":true,"requiresLowerAndUpperCompletedAnchors":true,"extrapolation":"refused","measuredModelSelection":"Median across available seeds at a batch shared across those seeds; choose the batch with maximum TPS"},"validation":{"scheme":"Leave one interior dual-seed model-size anchor out within each GPU/context curve; predict it from the nearest remaining lower and upper anchors","global":{"predictionCount":308,"medianAbsolutePercentError":2.096430759670309,"meanAbsolutePercentError":2.7915323785582253,"p90AbsolutePercentError":5.742545391121802,"maximumAbsolutePercentError":14.319880877757363,"meanSignedPercentError":-0.48755092264901567,"within10PercentPct":98.05194805194805,"within20PercentPct":100.0},"limitations":["Validation measures interpolation among OLM presets, not architecture-family shift.","The estimate does not predict batch size, MFU, or VRAM for the synthetic model.","Initialization, data input, evaluation, checkpointing, failures, and multi-GPU communication remain excluded."]},"gpus":{"NVIDIA RTX A6000":{"frontierAnchorCount":39,"contexts":{"512":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":64,"tps":37973.60698590911,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":32,"tps":25223.730414953858,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":32,"tps":19595.61098467291,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":32,"tps":15512.995759461111,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":16,"tps":11492.059154791357,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":16,"tps":8861.833806523216,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":8,"tps":4573.968322366779,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":2008824320},"1024":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":32,"tps":37319.62702394657,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":16,"tps":24708.23414491595,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":16,"tps":19241.366147121982,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":16,"tps":15229.7841503801,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":8,"tps":11324.09481096211,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":8,"tps":8729.714220483242,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":4,"tps":4525.980911108347,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":2008824320},"2048":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":16,"tps":36153.889191099996,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":8,"tps":23897.63098887112,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":8,"tps":18589.242744374766,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":8,"tps":14727.831498073352,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":4,"tps":10985.01868333221,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":4,"tps":8476.44921841934,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":2,"tps":4416.061425440939,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":2008824320},"4096":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":8,"tps":33479.69486804096,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":4,"tps":22116.694680082335,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":4,"tps":17289.342372637813,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":4,"tps":13727.100172296476,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":2,"tps":10312.154990738158,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":2,"tps":7993.637374830546,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":1,"tps":4228.087151961489,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":2008824320},"8192":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":4,"tps":29336.10234602692,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":2,"tps":19324.080956665206,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":2,"tps":15180.879179071642,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":2,"tps":12121.146687749824,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":1,"tps":9231.273301139248,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":1,"tps":7204.279693668515,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":995135232},"16384":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":2,"tps":23560.57247776167,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":1,"tps":15554.548551731003,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":1,"tps":12284.996006455338,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":1,"tps":9889.61876570026,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":505516800},"32768":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":1,"tps":16793.32778634305,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":150436608}},"validation":{"predictionCount":26,"medianAbsolutePercentError":2.351803693071519,"meanAbsolutePercentError":2.7590777194612297,"p90AbsolutePercentError":4.152084318009552,"maximumAbsolutePercentError":4.567120971049443,"meanSignedPercentError":-0.01727368442801482,"within10PercentPct":100.0,"within20PercentPct":100.0}},"NVIDIA GeForce RTX 4090":{"frontierAnchorCount":27,"contexts":{"512":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":16,"tps":62547.90415003331,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":16,"tps":40837.651779647495,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":16,"tps":31320.90280528467,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":8,"tps":22736.161232087597,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":8,"tps":17192.64615703041,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":4,"tps":11493.802305762994,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":995135232},"1024":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":8,"tps":60829.13608055089,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":8,"tps":39871.9466178241,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":8,"tps":30638.510849654678,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":4,"tps":22265.138416939444,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":4,"tps":16866.880408846373,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":2,"tps":11288.827750376655,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":995135232},"2048":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":4,"tps":58129.6974788426,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":4,"tps":38082.05736138195,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":4,"tps":29329.231722379507,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":2,"tps":21361.301847897495,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":2,"tps":16301.69409937058,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":1,"tps":10971.806835923799,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":995135232},"4096":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":2,"tps":52865.03648458059,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":2,"tps":34763.2334488467,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":2,"tps":26910.125261281402,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":1,"tps":19775.228345802607,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":1,"tps":15201.848877785105,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":707480064},"8192":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":1,"tps":45629.26328359115,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":1,"tps":29926.975337088603,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":1,"tps":23454.991487168794,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":348447744},"16384":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":1,"tps":35737.737411829934,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":150436608}},"validation":{"predictionCount":16,"medianAbsolutePercentError":1.8608490141633016,"meanAbsolutePercentError":2.400463922563436,"p90AbsolutePercentError":5.7513486052314375,"maximumAbsolutePercentError":5.855059447545274,"meanSignedPercentError":-0.6153844080756969,"within10PercentPct":100.0,"within20PercentPct":100.0}},"NVIDIA RTX 6000 Ada Generation":{"frontierAnchorCount":34,"contexts":{"512":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":16,"tps":60584.6164627444,"seeds":[11],"partial":true},{"modelKey":"250m","parameters":246397312,"batch":16,"tps":38310.58998401022,"seeds":[11],"partial":true},{"modelKey":"350m","parameters":348447744,"batch":16,"tps":29017.774931154043,"seeds":[11],"partial":true},{"modelKey":"500m","parameters":505516800,"batch":8,"tps":20811.930700927904,"seeds":[11],"partial":true},{"modelKey":"700m","parameters":707480064,"batch":16,"tps":15573.674637199874,"seeds":[11],"partial":true},{"modelKey":"1b","parameters":995135232,"batch":16,"tps":11452.92898342783,"seeds":[11],"partial":true},{"modelKey":"2b","parameters":2008824320,"batch":8,"tps":5905.209704489012,"seeds":[11],"partial":true}],"minimumParameters":150436608,"maximumParameters":2008824320},"1024":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":8,"tps":59070.32984903296,"seeds":[11],"partial":true},{"modelKey":"250m","parameters":246397312,"batch":8,"tps":37342.08391524828,"seeds":[11],"partial":true},{"modelKey":"350m","parameters":348447744,"batch":8,"tps":28446.323848609885,"seeds":[11],"partial":true},{"modelKey":"500m","parameters":505516800,"batch":8,"tps":20488.598139500908,"seeds":[11],"partial":true},{"modelKey":"700m","parameters":707480064,"batch":8,"tps":15363.178489629305,"seeds":[11],"partial":true},{"modelKey":"1b","parameters":995135232,"batch":8,"tps":11293.910310849065,"seeds":[11],"partial":true},{"modelKey":"2b","parameters":2008824320,"batch":4,"tps":5835.781747981443,"seeds":[11],"partial":true}],"minimumParameters":150436608,"maximumParameters":2008824320},"2048":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":4,"tps":57203.23564065523,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":4,"tps":36159.1813021175,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":4,"tps":27417.49818099003,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":4,"tps":19951.39531706755,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":4,"tps":14869.289843125745,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":4,"tps":10989.375746873942,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":2,"tps":5694.3950125487445,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":2008824320},"4096":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":2,"tps":51766.222681448555,"seeds":[11],"partial":true},{"modelKey":"250m","parameters":246397312,"batch":2,"tps":33094.10948413049,"seeds":[11],"partial":true},{"modelKey":"350m","parameters":348447744,"batch":1,"tps":25643.272484299985,"seeds":[11],"partial":true},{"modelKey":"500m","parameters":505516800,"batch":1,"tps":18770.85502013667,"seeds":[11],"partial":true},{"modelKey":"700m","parameters":707480064,"batch":2,"tps":14068.15529737064,"seeds":[11],"partial":true},{"modelKey":"1b","parameters":995135232,"batch":2,"tps":10458.004619147106,"seeds":[11],"partial":true},{"modelKey":"2b","parameters":2008824320,"batch":1,"tps":5509.341677528939,"seeds":[11],"partial":true}],"minimumParameters":150436608,"maximumParameters":2008824320},"8192":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":1,"tps":45711.114286020806,"seeds":[11],"partial":true},{"modelKey":"250m","parameters":246397312,"batch":1,"tps":29311.729525597613,"seeds":[11],"partial":true},{"modelKey":"350m","parameters":348447744,"batch":1,"tps":22617.920868629146,"seeds":[11],"partial":true},{"modelKey":"500m","parameters":505516800,"batch":1,"tps":16708.26568671067,"seeds":[11],"partial":true},{"modelKey":"700m","parameters":707480064,"batch":1,"tps":12831.911636327792,"seeds":[11],"partial":true},{"modelKey":"1b","parameters":995135232,"batch":1,"tps":9636.55652479585,"seeds":[11],"partial":true}],"minimumParameters":150436608,"maximumParameters":995135232}},"validation":{"predictionCount":5,"medianAbsolutePercentError":0.9946018027453318,"meanAbsolutePercentError":1.0775000272957813,"p90AbsolutePercentError":2.0756989960481365,"maximumAbsolutePercentError":2.7030825378614676,"meanSignedPercentError":0.0037329878488057133,"within10PercentPct":100.0,"within20PercentPct":100.0}},"NVIDIA GeForce RTX 5090":{"frontierAnchorCount":31,"contexts":{"512":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":32,"tps":93684.24096240377,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":16,"tps":59118.34094167946,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":16,"tps":44165.536469843566,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":16,"tps":35299.29321040954,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":16,"tps":26576.954824061613,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":8,"tps":18110.420437619243,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":995135232},"1024":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":16,"tps":91327.64212261455,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":8,"tps":57515.90319071589,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":8,"tps":43121.35359298127,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":8,"tps":34533.06030331098,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":8,"tps":26067.72094098488,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":4,"tps":17787.91298614558,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":995135232},"2048":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":8,"tps":87013.34577039396,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":4,"tps":54840.98949375695,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":4,"tps":41288.75854525097,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":4,"tps":33158.335834707905,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":4,"tps":25129.173754423384,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":2,"tps":17194.590858293344,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":995135232},"4096":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":4,"tps":79499.58874816081,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":2,"tps":49808.72009891279,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":2,"tps":37845.22636132228,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":2,"tps":30596.656203583938,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":2,"tps":23400.03250222741,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":1,"tps":16171.027839734816,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":995135232},"8192":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":2,"tps":68081.75089527536,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":2,"tps":43042.2954581962,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":1,"tps":32771.08095373423,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":1,"tps":26716.801588241513,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":1,"tps":20692.845319473316,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":707480064},"16384":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":1,"tps":53382.246827019684,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":1,"tps":34174.60416689911,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":246397312}},"validation":{"predictionCount":19,"medianAbsolutePercentError":4.145508851119583,"meanAbsolutePercentError":3.8250213008117315,"p90AbsolutePercentError":4.693327779529974,"maximumAbsolutePercentError":4.747709501738173,"meanSignedPercentError":-0.25552475176390255,"within10PercentPct":100.0,"within20PercentPct":100.0}},"NVIDIA A100 80GB PCIe":{"frontierAnchorCount":49,"contexts":{"512":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":64,"tps":74602.70513502024,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":64,"tps":49994.19964205101,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":64,"tps":39117.3571418394,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":64,"tps":30950.228537889117,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":32,"tps":23229.128733985053,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":32,"tps":18069.502684639556,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":16,"tps":10170.25980027553,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":8,"tps":5139.333740361808,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"1024":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":32,"tps":71721.12940395741,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":32,"tps":48607.1898806909,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":32,"tps":38102.46387005976,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":32,"tps":30196.409965475388,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":16,"tps":22738.393548234526,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":16,"tps":17774.561867898527,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":8,"tps":10025.762084257229,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":4,"tps":5070.573987487038,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"2048":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":16,"tps":68757.5947315981,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":16,"tps":46133.39673857178,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":16,"tps":36133.43218329415,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":16,"tps":28762.432391752558,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":8,"tps":21739.401615201976,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":8,"tps":17044.018663269406,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":4,"tps":9718.23274048148,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":2,"tps":4955.745887835652,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"4096":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":8,"tps":61850.90931413724,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":8,"tps":41671.83307504239,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":8,"tps":32929.2443195915,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":8,"tps":26256.159682480553,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":4,"tps":19994.15055470971,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":4,"tps":15753.620832494002,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":2,"tps":9212.167580496412,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":1,"tps":4748.075938466991,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"8192":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":4,"tps":52183.4350009785,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":4,"tps":35217.464357022356,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":4,"tps":27889.48049424085,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":4,"tps":22372.663448935746,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":2,"tps":17250.75264149651,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":2,"tps":13676.95193437717,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":1,"tps":8264.330220975113,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":2008824320},"16384":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":2,"tps":40597.624438093386,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":2,"tps":26919.72710822182,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":2,"tps":21460.069309512135,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":2,"tps":17287.539934708366,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":1,"tps":13651.234999313256,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":1,"tps":10918.919982190204,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":995135232},"32768":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":1,"tps":28016.56580648503,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":1,"tps":18370.617936954834,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":1,"tps":14744.785377968954,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":1,"tps":11977.914732514502,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":505516800}},"validation":{"predictionCount":35,"medianAbsolutePercentError":2.087798274917252,"meanAbsolutePercentError":2.6482695442034037,"p90AbsolutePercentError":5.340900508132841,"maximumAbsolutePercentError":6.672937277124047,"meanSignedPercentError":-0.3772717340497413,"within10PercentPct":100.0,"within20PercentPct":100.0}},"NVIDIA RTX PRO 6000 Blackwell Server Edition":{"frontierAnchorCount":50,"contexts":{"512":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":32,"tps":105487.48125689088,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":32,"tps":71817.37971261576,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":32,"tps":56894.872033455555,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":16,"tps":43449.181193182034,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":16,"tps":32678.46265577907,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":16,"tps":25554.305358507972,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":16,"tps":14469.301860774893,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":8,"tps":7382.144190540364,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"1024":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":16,"tps":105326.47558462239,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":16,"tps":70648.55000176342,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":16,"tps":56083.296598682035,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":8,"tps":42740.990937961644,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":8,"tps":32214.29514983353,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":8,"tps":25252.520528657707,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":8,"tps":14362.791549440415,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":4,"tps":7337.016491226279,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"2048":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":8,"tps":101657.0747283123,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":8,"tps":68197.81811601312,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":8,"tps":54201.101839967596,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":4,"tps":41283.374058431335,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":4,"tps":31314.62291915714,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":4,"tps":24587.737967686804,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":8,"tps":14646.715609806552,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":2,"tps":7217.362406804861,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"4096":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":4,"tps":94764.51109743607,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":4,"tps":63590.29116061887,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":4,"tps":50761.4286376642,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":2,"tps":38645.45273706463,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":2,"tps":29496.639960228873,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":2,"tps":23238.42579180834,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":2,"tps":13484.804715014474,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":1,"tps":6992.177850260997,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"8192":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":2,"tps":83740.92153558889,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":2,"tps":56233.541687973964,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":2,"tps":44932.984013320194,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":1,"tps":34603.70093779536,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":1,"tps":26688.317763977946,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":1,"tps":21147.419194763428,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":1,"tps":12494.417549156624,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":2008824320},"16384":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":1,"tps":68778.31880048677,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":1,"tps":46097.54240827837,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":1,"tps":36951.29413157355,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":1,"tps":28363.375250073765,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":2,"tps":22619.581073568217,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":1,"tps":18007.782078923516,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":995135232},"32768":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":2,"tps":49626.18328090579,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":1,"tps":32848.99376869881,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":1,"tps":26520.60392143711,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":1,"tps":21516.354880814004,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":1,"tps":17408.493124128192,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":707480064}},"validation":{"predictionCount":36,"medianAbsolutePercentError":1.8637699076886505,"meanAbsolutePercentError":2.3817754229899393,"p90AbsolutePercentError":5.105336175493186,"maximumAbsolutePercentError":9.599286439935407,"meanSignedPercentError":-0.47786765934720277,"within10PercentPct":100.0,"within20PercentPct":100.0}},"NVIDIA H100 80GB HBM3":{"frontierAnchorCount":49,"contexts":{"512":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":64,"tps":151610.95094063462,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":64,"tps":104029.38796097372,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":64,"tps":84269.36966203585,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":64,"tps":68539.52093726992,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":32,"tps":52836.11279899211,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":32,"tps":42071.71797078621,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":16,"tps":24110.24492712824,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":8,"tps":12476.269612662483,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"1024":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":32,"tps":147577.36075110853,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":32,"tps":101099.35818877764,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":32,"tps":81942.96274543274,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":32,"tps":66703.50389364499,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":16,"tps":51489.00871991577,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":16,"tps":41043.492189165845,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":8,"tps":23657.891546945255,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":4,"tps":12292.299966867224,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"2048":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":16,"tps":139706.04386329444,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":16,"tps":95631.56917040798,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":16,"tps":77478.37249399451,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":16,"tps":63145.15692877116,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":8,"tps":48892.16240084113,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":8,"tps":39017.141723338835,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":4,"tps":22708.321172163596,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":2,"tps":11878.135470235095,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"4096":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":8,"tps":125991.65211796765,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":8,"tps":85914.73151578513,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":8,"tps":69794.55079171214,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":8,"tps":57008.37510355465,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":4,"tps":44220.02874116771,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":4,"tps":35561.51889683539,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":2,"tps":21071.536881729466,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":1,"tps":11241.313747427193,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"8192":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":4,"tps":105525.90434378124,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":4,"tps":71672.39335637711,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":4,"tps":58251.41269121611,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":4,"tps":47733.02307635457,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":2,"tps":37352.785684543436,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":2,"tps":30194.4227916865,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":1,"tps":18561.854204000323,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":2008824320},"16384":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":2,"tps":80428.7915247498,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":2,"tps":54203.8889963037,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":2,"tps":44089.87260300973,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":2,"tps":36219.322550293655,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":1,"tps":28951.828185757367,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":1,"tps":23465.96683528873,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":995135232},"32768":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":1,"tps":54551.34299247776,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":1,"tps":36456.654072203426,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":1,"tps":29650.14534090385,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":1,"tps":24422.221444648505,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":505516800}},"validation":{"predictionCount":35,"medianAbsolutePercentError":2.712343840573328,"meanAbsolutePercentError":2.8316360171101156,"p90AbsolutePercentError":5.163611177261711,"maximumAbsolutePercentError":5.754893266239581,"meanSignedPercentError":-0.32537844973750213,"within10PercentPct":100.0,"within20PercentPct":100.0}},"NVIDIA H200":{"frontierAnchorCount":56,"contexts":{"512":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":128,"tps":180347.54889587476,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":128,"tps":122920.62177856671,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":128,"tps":97295.99941836175,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":64,"tps":75985.67285341932,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":64,"tps":61182.6600054468,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":64,"tps":47740.67746869053,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":32,"tps":27745.248130385855,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":16,"tps":15243.625915880242,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":4,"tps":6146.694697319001,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"1024":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":64,"tps":174615.32933460455,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":64,"tps":118805.53941349333,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":64,"tps":94182.2039494935,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":32,"tps":73529.13938092274,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":32,"tps":59336.931907655635,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":32,"tps":46399.73213940572,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":16,"tps":27123.027833521093,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":8,"tps":14996.732439374835,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":2,"tps":6070.791722177572,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"2048":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":32,"tps":163858.07131755044,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":32,"tps":111331.80277302605,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":32,"tps":88361.82054660862,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":16,"tps":69185.30462049143,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":16,"tps":55921.71514490085,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":16,"tps":43902.57845736956,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":8,"tps":25957.230482553925,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":4,"tps":14482.344820516191,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":1,"tps":5952.803097026264,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"4096":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":16,"tps":145775.20163147504,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":16,"tps":98650.91038693838,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":16,"tps":78455.28584810357,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":8,"tps":61837.48482879089,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":8,"tps":50172.17680305708,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":8,"tps":39110.254149310014,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":4,"tps":23905.375702354875,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":2,"tps":13568.859167935057,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"8192":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":8,"tps":119507.78383449535,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":8,"tps":80408.19313565397,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":8,"tps":64433.597499236996,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":4,"tps":51141.28272270268,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":4,"tps":41632.026029710294,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":4,"tps":33148.94244543079,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":2,"tps":20676.58637196062,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":1,"tps":12075.250468096256,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"16384":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":4,"tps":88384.27325452096,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":4,"tps":59083.96061465266,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":4,"tps":47451.13063650114,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":2,"tps":38166.763111301385,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":2,"tps":31326.432586065595,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":2,"tps":25149.030944826736,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":1,"tps":16585.028942794073,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":2008824320},"32768":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":2,"tps":58124.38810740897,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":2,"tps":38594.905474847634,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":2,"tps":31178.044336483465,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":1,"tps":25321.820187318233,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":1,"tps":20897.14719947258,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":1,"tps":16905.197716587296,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":995135232}},"validation":{"predictionCount":42,"medianAbsolutePercentError":1.0307320749336601,"meanAbsolutePercentError":2.382368472443025,"p90AbsolutePercentError":4.10527673197478,"maximumAbsolutePercentError":14.319880877757363,"meanSignedPercentError":-1.0615483587474523,"within10PercentPct":92.85714285714286,"within20PercentPct":100.0}},"NVIDIA B200":{"frontierAnchorCount":60,"contexts":{"512":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":256,"tps":240414.0461881712,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":128,"tps":158348.88684645877,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":128,"tps":130210.15705399853,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":128,"tps":108189.43950285445,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":128,"tps":89102.43941470925,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":64,"tps":68565.73717943975,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":64,"tps":44866.96849258431,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":32,"tps":25862.362386048142,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":16,"tps":13691.751664353724,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"1024":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":128,"tps":230886.89631543745,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":64,"tps":152039.0524378501,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":64,"tps":125118.72991391704,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":64,"tps":103889.37961680413,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":64,"tps":85655.13113080667,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":32,"tps":65830.7044568414,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":32,"tps":43608.86407261456,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":16,"tps":25191.344787094786,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":8,"tps":13422.29103061408,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"2048":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":64,"tps":213575.04399362917,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":32,"tps":140055.57101578967,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":32,"tps":115400.54970827686,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":32,"tps":95798.6780661942,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":32,"tps":79291.85343366751,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":16,"tps":61294.06622840036,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":16,"tps":41068.092218032485,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":8,"tps":24003.630038237152,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":4,"tps":12853.909648964625,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"4096":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":32,"tps":186070.58960697224,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":16,"tps":122199.91705226383,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":16,"tps":100494.39165820224,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":16,"tps":83383.09511136063,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":16,"tps":68976.93648092463,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":8,"tps":53618.79771063005,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":8,"tps":36592.99814124362,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":4,"tps":21810.77640533662,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":2,"tps":11977.004159637283,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"8192":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":16,"tps":147599.1095375495,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":8,"tps":96878.3701496892,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":8,"tps":79478.80688404571,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":8,"tps":66016.56842063848,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":8,"tps":54718.06565502567,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":4,"tps":42967.167777902156,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":4,"tps":30070.290877725958,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":2,"tps":18356.49004561978,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":1,"tps":10422.420819554492,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"16384":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":8,"tps":104194.54251395656,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":4,"tps":68322.5027956648,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":4,"tps":55959.31688571662,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":4,"tps":46473.88979857553,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":4,"tps":38626.665499171984,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":2,"tps":30736.03112032974,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":2,"tps":22202.79247450845,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":1,"tps":13993.223723270428,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392},"32768":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":4,"tps":66283.42515233991,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":2,"tps":43393.23796865513,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":2,"tps":35436.101894765045,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":2,"tps":29342.662453882353,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":2,"tps":24425.8822677524,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":1,"tps":19673.530653820082,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":1,"tps":14663.731819799319,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":2008824320}},"validation":{"predictionCount":46,"medianAbsolutePercentError":3.850397206010742,"meanAbsolutePercentError":3.677199867885643,"p90AbsolutePercentError":6.776777591782079,"maximumAbsolutePercentError":7.060771035197444,"meanSignedPercentError":-0.11327489869223456,"within10PercentPct":100.0,"within20PercentPct":100.0}},"NVIDIA B300 SXM6 AC":{"frontierAnchorCount":62,"contexts":{"512":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":256,"tps":244426.60397349685,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":256,"tps":167668.59844259324,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":256,"tps":137170.4374736513,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":128,"tps":108185.66919720893,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":128,"tps":89335.80539700901,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":128,"tps":71833.40409613764,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":64,"tps":42600.70253899779,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":64,"tps":25628.697756362857,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":32,"tps":12338.966029140676,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"1024":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":128,"tps":234699.49003352463,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":128,"tps":160825.81660632417,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":128,"tps":131566.3400125942,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":64,"tps":103951.57508126489,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":64,"tps":85890.8989818762,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":64,"tps":69117.21772459729,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":32,"tps":41427.23343726953,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":32,"tps":25043.183773326797,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":16,"tps":12144.33371625455,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"2048":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":64,"tps":217433.12464187917,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":64,"tps":148584.81644025864,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":64,"tps":121515.35605577381,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":32,"tps":96290.73876641985,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":32,"tps":79608.26720024087,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":32,"tps":64153.67715820215,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":16,"tps":39067.409642456354,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":16,"tps":23828.220736276424,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":8,"tps":11724.40056777219,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"4096":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":32,"tps":189129.17382159331,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":32,"tps":128628.16786770508,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":32,"tps":105155.02640650692,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":16,"tps":83723.18402372937,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":16,"tps":69335.9370524845,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":16,"tps":56096.40425707761,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":8,"tps":35119.26385592564,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":8,"tps":21693.918591205227,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":4,"tps":11043.39392923747,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"8192":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":16,"tps":149999.0743961719,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":16,"tps":101379.44282176913,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":16,"tps":82870.58010358897,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":8,"tps":66397.42417162463,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":8,"tps":55116.866591655504,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":8,"tps":44775.894542056354,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":4,"tps":29179.32234449318,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":4,"tps":18403.57216925877,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":2,"tps":9773.602696522248,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"16384":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":8,"tps":105848.1371485323,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":8,"tps":71053.2375807256,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":8,"tps":57993.38920916847,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":4,"tps":46885.369136810186,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":4,"tps":38975.38198861583,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":4,"tps":31836.26286962047,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":2,"tps":21876.933877476404,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":2,"tps":14136.211884795748,"seeds":[11,22],"partial":false},{"modelKey":"8b","parameters":7983108096,"batch":1,"tps":7962.276142785917,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":7983108096},"32768":{"anchors":[{"modelKey":"150m","parameters":150436608,"batch":4,"tps":67285.53395436739,"seeds":[11,22],"partial":false},{"modelKey":"250m","parameters":246397312,"batch":4,"tps":44776.47436910801,"seeds":[11,22],"partial":false},{"modelKey":"350m","parameters":348447744,"batch":4,"tps":36486.06013956273,"seeds":[11,22],"partial":false},{"modelKey":"500m","parameters":505516800,"batch":2,"tps":29717.83720417207,"seeds":[11,22],"partial":false},{"modelKey":"700m","parameters":707480064,"batch":2,"tps":24734.467968081153,"seeds":[11,22],"partial":false},{"modelKey":"1b","parameters":995135232,"batch":2,"tps":20281.11026574926,"seeds":[11,22],"partial":false},{"modelKey":"2b","parameters":2008824320,"batch":1,"tps":14684.819325923765,"seeds":[11,22],"partial":false},{"modelKey":"4b","parameters":3999611392,"batch":1,"tps":9702.810065860152,"seeds":[11,22],"partial":false}],"minimumParameters":150436608,"maximumParameters":3999611392}},"validation":{"predictionCount":48,"medianAbsolutePercentError":1.1617571442699037,"meanAbsolutePercentError":2.600715513293064,"p90AbsolutePercentError":7.0301318327851305,"maximumAbsolutePercentError":10.441782338053486,"meanSignedPercentError":-0.9050509152291365,"within10PercentPct":93.75,"within20PercentPct":100.0}}}}}