Qwen3.6-27B-uncensored-heretic-v2-Native-MTP-Preserved-PRISMAQUANT-PRISMASCOUT-6bit / mixed_native_manifest.json
| { | |
| "source_model": "/home/jurajpetrik/.cache/huggingface/hub/models--llmfan46--Qwen3.6-27B-uncensored-heretic-v2-Native-MTP-Preserved/snapshots/4928dd14830859583b97f35dfaba9018f7310b3e", | |
| "source_recipe": "./dq-runs/qwen36/artifacts/layer_config.json", | |
| "format_histogram": { | |
| "head_passthrough/BF16": 3, | |
| "linear/NVFP4_PRODUCTION_CACHE": 243, | |
| "linear/MXFP8_E4M3_PRODUCTION_CACHE": 128, | |
| "linear/BF16": 125, | |
| "layer_passthrough/BF16": 352, | |
| "mtp_linear/BF16": 8, | |
| "mtp_passthrough/BF16": 7 | |
| }, | |
| "n_assignment_entries": 614, | |
| "runtime_coercions": [], | |
| "bf16_audit": { | |
| "counts": { | |
| "allocator_selected_bf16_alternate_8bit_legal": 119, | |
| "allocator_selected_bf16_mxfp8_legal": 124 | |
| }, | |
| "entries": [ | |
| { | |
| "name": "model.layers.0.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.0.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.1.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.1.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.10.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.10.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.11.self_attn.k_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.11.self_attn.q_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 12288, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.11.self_attn.v_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.12.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.12.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.13.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.13.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.14.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.14.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.15.self_attn.k_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.15.self_attn.q_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 12288, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.15.self_attn.v_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.16.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.16.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.17.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.17.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.17.linear_attn.in_proj_qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 10240, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.17.linear_attn.in_proj_z", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 6144, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.18.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.18.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.18.linear_attn.in_proj_qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 10240, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.18.linear_attn.in_proj_z", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 6144, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.19.self_attn.k_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.19.self_attn.q_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 12288, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.19.self_attn.v_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.2.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.2.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.20.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.20.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.20.linear_attn.in_proj_qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 10240, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.20.linear_attn.in_proj_z", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 6144, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.21.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.21.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.22.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.22.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.23.self_attn.k_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.23.self_attn.q_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 12288, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.23.self_attn.v_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.24.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.24.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.25.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.25.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.26.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.26.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.27.self_attn.k_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.27.self_attn.q_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 12288, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.27.self_attn.v_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.28.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.28.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.29.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.29.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.3.self_attn.k_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.3.self_attn.q_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 12288, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.3.self_attn.v_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.30.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.30.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.31.self_attn.k_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.31.self_attn.q_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 12288, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.31.self_attn.v_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.32.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.32.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.33.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.33.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.34.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.34.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.36.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.36.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.37.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.37.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.38.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.38.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.4.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.4.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.40.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.40.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.41.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.41.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.42.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.42.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.44.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.44.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.45.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.45.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.46.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.46.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.48.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.48.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.49.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.49.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.5.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.5.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.50.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.50.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.52.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.52.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.53.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.53.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.54.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.54.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.56.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.56.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.6.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.6.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.6.linear_attn.in_proj_qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 10240, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.6.linear_attn.in_proj_z", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 6144, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.60.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.60.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.61.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.61.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.62.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.62.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.63.mlp.down_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 5120, | |
| 17408 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.7.self_attn.k_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.7.self_attn.q_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 12288, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.7.self_attn.v_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.8.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.8.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.9.linear_attn.in_proj_a", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.layers.9.linear_attn.in_proj_b", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 48, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "kernel_shape", | |
| "detail": "vLLM/FlashInfer MXFP8 GEMM requires K >= 128, N >= 128, and K divisible by the 32-value MX block size; swizzled scale layouts pad N internally (out_features=48, in_features=5120)" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.0.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.0.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.0.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.0.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.1.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.1.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.1.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.1.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.10.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.10.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.10.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.10.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.11.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.11.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.11.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.11.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.12.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.12.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.12.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.12.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.13.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.13.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.13.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.13.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.14.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.14.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.14.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.14.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.15.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.15.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.15.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.15.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.16.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.16.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.16.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.16.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.17.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.17.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.17.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.17.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.18.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.18.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.18.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.18.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.19.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.19.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.19.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.19.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.2.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.2.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.2.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.2.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.20.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.20.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.20.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.20.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.21.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.21.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.21.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.21.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.22.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.22.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.22.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.22.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.23.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.23.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.23.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.23.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.24.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.24.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.24.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.24.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.25.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.25.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.25.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.25.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.26.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.26.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.26.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.26.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.3.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.3.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.3.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.3.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.4.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.4.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.4.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.4.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.5.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.5.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.5.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.5.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.6.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.6.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.6.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.6.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.7.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.7.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.7.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.7.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.8.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.8.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.8.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.8.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.9.attn.proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1152, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.9.attn.qkv", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 3456, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.9.mlp.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4304, | |
| 1152 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.blocks.9.mlp.linear_fc2", | |
| "reason": "allocator_selected_bf16_alternate_8bit_legal", | |
| "shape": [ | |
| 1152, | |
| 4304 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": false, | |
| "reason": "group_divisibility", | |
| "detail": "group_size=32 does not divide in_features=4304" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.merger.linear_fc1", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 4608, | |
| 4608 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "model.visual.merger.linear_fc2", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 5120, | |
| 4608 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "mtp.fc", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 5120, | |
| 10240 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "mtp.layers.0.mlp.down_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 5120, | |
| 17408 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "mtp.layers.0.mlp.gate_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 17408, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "mtp.layers.0.mlp.up_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 17408, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "mtp.layers.0.self_attn.k_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "mtp.layers.0.self_attn.o_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 5120, | |
| 6144 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "mtp.layers.0.self_attn.q_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 12288, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| }, | |
| { | |
| "name": "mtp.layers.0.self_attn.v_proj", | |
| "reason": "allocator_selected_bf16_mxfp8_legal", | |
| "shape": [ | |
| 1024, | |
| 5120 | |
| ], | |
| "eight_bit_candidates": { | |
| "MXFP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "MXFP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| }, | |
| "FP8_E4M3": { | |
| "legal": true, | |
| "reason": null, | |
| "detail": "" | |
| }, | |
| "FP8_E5M2": { | |
| "legal": false, | |
| "reason": "profile_mismatch", | |
| "detail": "This format is not enabled for dense Linears in the selected vLLM serving profile" | |
| } | |
| } | |
| } | |
| ], | |
| "target_profile": "vllm_packed_moe" | |
| }, | |
| "ignore": [ | |
| "lm_head" | |
| ] | |
| } |