Add FP8 validation evidence

validation/performance_metrics.json

@@ -0,0 +1,47 @@
+{
+  "target": "FP8",
+  "model": "/mnt/c/Users/chunk/OneDrive/Documents/quant3.8/qwen38-cuda-quants/outputs/Qwen3.8-27B-FP8",
+  "created_utc": "2026-08-18T08:12:26+00:00",
+  "smoke_test": false,
+  "execution_mode": "RTX 4070 + CPU offload/auto map",
+  "hardware": "NVIDIA GeForce RTX 4070 12 GB (not 4070 Ti)",
+  "device_map_summary": {
+    "cpu": 59,
+    "cuda:0": 9
+  },
+  "repetitions": 1,
+  "prompt_tokens": 8,
+  "requested_decode_tokens": 1,
+  "warmup_tokens": 1,
+  "cold_load_seconds": 135.15700745499998,
+  "warm_load_seconds": null,
+  "single_load_memory_safe_mode": true,
+  "ttft_seconds_median": 246.20996074599998,
+  "prefill_tok_s_median": 0.049479823627132595,
+  "decode_tok_s_median": 0.023630125161260673,
+  "end_to_end_generated_tok_s_median": 0.004901938544242578,
+  "median_generated_tokens": 1.0,
+  "peak_vram_gib": 11.6689453125,
+  "peak_vram_source": "nvidia-smi device-wide memory.used sampled every 0.5 seconds",
+  "baseline_device_memory_gib": 1.2080078125,
+  "device_memory_total_gib": 11.994140625,
+  "torch_allocator_peak_gib": 10.919785499572754,
+  "device_memory_sampler_error": null,
+  "process_rss_gib": 47.185646057128906,
+  "system_used_gib_at_end": 48.21872329711914,
+  "raw": {
+    "prefill_seconds": [
+      161.68206378999997
+    ],
+    "ttft_seconds": [
+      246.20996074599998
+    ],
+    "generation_seconds": [
+      204.00092554700007
+    ],
+    "generated_tokens": [
+      1
+    ]
+  },
+  "passed": true
+}

validation/validation_metrics.json

@@ -0,0 +1,28 @@
+{
+  "target": "FP8",
+  "reference": "/mnt/c/Users/chunk/OneDrive/Documents/quant3.8/qwen38-cuda-quants/hf-cache/source/Qwen3.8-27B--1d4bf0f2ff60",
+  "quant": "/mnt/c/Users/chunk/OneDrive/Documents/quant3.8/qwen38-cuda-quants/outputs/Qwen3.8-27B-FP8",
+  "created_utc": "2026-08-17T03:15:49+00:00",
+  "smoke_test": false,
+  "thresholds": {
+    "max_mean_kl": 0.2,
+    "max_p95_kl": 0.8,
+    "min_top1": 0.6,
+    "min_top5_overlap": 0.6,
+    "min_logit_cosine": 0.9,
+    "max_ppl_delta_percent": 50.0
+  },
+  "token_positions": 64,
+  "kl_direction": "KL(P_BF16 || P_QUANT)",
+  "kl_mean": 0.005752270109951496,
+  "kl_median": 0.0024972036480903625,
+  "kl_p95": 0.01725013740360737,
+  "top1_agreement": 0.96875,
+  "top5_set_overlap": 0.9750000238418579,
+  "reference_nll": 3.4673221111297607,
+  "quant_nll": 3.4618234634399414,
+  "nll_delta": -0.005498647689819336,
+  "ppl_delta_percent": -0.5483557797280714,
+  "logit_cosine_similarity": 0.9966187477111816,
+  "passed": true
+}

validation/validation_report.json

@@ -0,0 +1,330 @@
+{
+  "target": "FP8",
+  "artifact": "/mnt/c/Users/chunk/OneDrive/Documents/quant3.8/qwen38-cuda-quants/outputs/Qwen3.8-27B-FP8",
+  "started_utc": "2026-08-18T06:44:09+00:00",
+  "smoke_test": false,
+  "small_architecture_test": false,
+  "resume_evidence": {
+    "path": "/mnt/c/Users/chunk/OneDrive/Documents/quant3.8/qwen38-cuda-quants/logs/full_fp8_validate_resume3.console.log",
+    "sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c",
+    "passing_generation_cases": [
+      "coding",
+      "debugging",
+      "eos",
+      "factual",
+      "json",
+      "long_context",
+      "math",
+      "multilingual",
+      "multiturn",
+      "thinking_off"
+    ],
+    "passing_tool_cases": [
+      "chained",
+      "missing_required",
+      "nested",
+      "weather"
+    ]
+  },
+  "checks": {
+    "config": {
+      "passed": true,
+      "class": "Qwen3_5Config",
+      "model_type": "qwen3_5",
+      "language_layers": 64,
+      "advertised_context_length": 262144,
+      "quantization_method": "fp8",
+      "expected_quantization_method": "fp8",
+      "quantization_config": {
+        "bits": 8,
+        "checkpoint_format": "float8_e4m3fn",
+        "desc_act": false,
+        "format": "float8_e4m3fn",
+        "group_size": -1,
+        "lm_head": false,
+        "meta": {
+          "act_group_aware": null,
+          "auto_forward_data_parallel": true,
+          "damp_auto_increment": null,
+          "damp_percent": null,
+          "dense_vram_strategy": "exclusive",
+          "dense_vram_strategy_devices": null,
+          "fallback": {
+            "smooth": null,
+            "strategy": "rtn",
+            "threshold": "0.5%"
+          },
+          "foem": null,
+          "gc_mode": "interval",
+          "gptaq": null,
+          "moe_vram_strategy": "exclusive",
+          "moe_vram_strategy_devices": null,
+          "mse": null,
+          "offload_to_disk": false,
+          "offload_to_disk_path": null,
+          "pack_impl": "cpu",
+          "quantizer": [
+            "gptqmodel:7.3.4"
+          ],
+          "static_groups": null,
+          "true_sequential": true,
+          "uri": "https://github.com/modelcloud/gptqmodel",
+          "wait_for_submodule_finalizers": false
+        },
+        "method": "fp8",
+        "pack_dtype": "int32",
+        "quant_method": "fp8",
+        "weight_scale_method": "row",
+        "weight_scale_semantics": "inverse"
+      }
+    },
+    "processor": {
+      "passed": true,
+      "class": "Qwen3VLProcessor",
+      "chat_template_present": true,
+      "special_token_ids": {
+        "image_token_id": 248056,
+        "video_token_id": 248057,
+        "vision_start_token_id": 248053,
+        "vision_end_token_id": 248054
+      }
+    },
+    "chat_template": {
+      "present": true,
+      "sha256": "c3cf9e34abf4f9e36c2d72165aa9c132d3e2a725b6c2586aaa3a8af9d7a81041",
+      "matches_reference": true,
+      "passed": true
+    },
+    "inherited_metadata": {
+      "passed": true,
+      "files_checked": [
+        "generation_config.json",
+        "tokenizer.json",
+        "tokenizer_config.json",
+        "vocab.json",
+        "merges.txt",
+        "chat_template.jinja",
+        "preprocessor_config.json",
+        "video_preprocessor_config.json",
+        "processor_config.json",
+        "LICENSE"
+      ],
+      "missing": [],
+      "mismatched": []
+    },
+    "tensors": {
+      "files": [
+        {
+          "name": "model-00001-of-00008.safetensors",
+          "bytes": 3516829256
+        },
+        {
+          "name": "model-00002-of-00008.safetensors",
+          "bytes": 4268842872
+        },
+        {
+          "name": "model-00003-of-00008.safetensors",
+          "bytes": 4270471840
+        },
+        {
+          "name": "model-00004-of-00008.safetensors",
+          "bytes": 4280916456
+        },
+        {
+          "name": "model-00005-of-00008.safetensors",
+          "bytes": 4265221120
+        },
+        {
+          "name": "model-00006-of-00008.safetensors",
+          "bytes": 4212755176
+        },
+        {
+          "name": "model-00007-of-00008.safetensors",
+          "bytes": 4270471904
+        },
+        {
+          "name": "model-00008-of-00008.safetensors",
+          "bytes": 2166188208
+        }
+      ],
+      "tensor_count": 1599,
+      "float_tensors_scanned": 1599,
+      "nan_count": 0,
+      "inf_count": 0,
+      "meta_tensor_count": 0,
+      "has_language": true,
+      "has_vision": true,
+      "has_lm_head": true,
+      "has_mtp": true,
+      "quantized_weight_tensors": 0,
+      "lm_head_satisfied_by_tied_embeddings": false,
+      "packed_weights_required": false,
+      "passed": true
+    },
+    "model_load": {
+      "passed": true,
+      "loader": "GPTQModel",
+      "model_class": "Qwen3_5ForConditionalGeneration",
+      "device_map_summary": {
+        "cpu": 59,
+        "cuda:0": 9
+      }
+    },
+    "generation": {
+      "passed": true,
+      "cases": [
+        {
+          "name": "factual",
+          "prompt": "Answer in one short sentence: What is the capital of France?",
+          "case_passed": true,
+          "new_tokens": 8,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        },
+        {
+          "name": "math",
+          "prompt": "Compute 17 x 23. Give only the integer.",
+          "case_passed": true,
+          "new_tokens": 4,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        },
+        {
+          "name": "coding",
+          "prompt": "Write a Python function is_even(n) in at most three lines.",
+          "case_passed": true,
+          "new_tokens": 21,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        },
+        {
+          "name": "debugging",
+          "prompt": "Fix this Python expression and explain briefly: len(3).",
+          "case_passed": true,
+          "new_tokens": 96,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        },
+        {
+          "name": "multiturn",
+          "prompt": "The secret word is cobalt. What secret word did I just give you?",
+          "case_passed": true,
+          "new_tokens": 3,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        },
+        {
+          "name": "json",
+          "prompt": "Return only valid JSON with keys \"status\" and \"count\"; use status \"ok\" and count 3.",
+          "case_passed": true,
+          "new_tokens": 13,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        },
+        {
+          "name": "multilingual",
+          "prompt": "Responde en espanol con una frase: De que color es el cielo despejado?",
+          "case_passed": true,
+          "new_tokens": 10,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        },
+        {
+          "name": "thinking_off",
+          "prompt": "Give only the final answer, without reasoning: 9 + 14 = ?",
+          "case_passed": true,
+          "new_tokens": 3,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        },
+        {
+          "name": "long_context",
+          "prompt": "Recover the codeword from the long context and return only that word.",
+          "case_passed": true,
+          "new_tokens": 4,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        },
+        {
+          "name": "eos",
+          "prompt": "Reply with exactly: done",
+          "case_passed": true,
+          "new_tokens": 2,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        }
+      ]
+    },
+    "tools": {
+      "passed": true,
+      "cases": [
+        {
+          "case": "weather",
+          "prompt": "Use the weather tool for Chicago with unit celsius. Return only the tool call.",
+          "schema_passed": true,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        },
+        {
+          "case": "nested",
+          "prompt": "Use the route tool with JSON arguments origin=A, destination=B, options={avoid_tolls:true}. Return only the tool call.",
+          "schema_passed": true,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        },
+        {
+          "case": "chained",
+          "prompt": "Call lookup_user for id 7, then call list_orders using that user. Return the two tool calls.",
+          "schema_passed": true,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        },
+        {
+          "case": "missing_required",
+          "prompt": "Before calling book_trip, state which required date is missing. Do not invent it.",
+          "schema_passed": true,
+          "resumed_from_evidence": true,
+          "evidence_sha256": "0f648b7067b40f85ea45320b1d841c38a79bbf11f022f6e3eb4a974e57072e8c"
+        }
+      ]
+    },
+    "vision": {
+      "text": "blue red green B 7",
+      "prompt_tokens": 106,
+      "new_tokens": 7,
+      "input_shapes": {
+        "input_ids": [
+          1,
+          106
+        ],
+        "attention_mask": [
+          1,
+          106
+        ],
+        "mm_token_type_ids": [
+          1,
+          106
+        ],
+        "pixel_values": [
+          280,
+          1536
+        ],
+        "image_grid_thw": [
+          1,
+          3
+        ]
+      },
+      "contains_all_shape_colors": true,
+      "contains_tallest_bar_fact": true,
+      "passed": true
+    }
+  },
+  "passed": true,
+  "finished_utc": "2026-08-18T07:45:36+00:00",
+  "elapsed_seconds": 3688.98,
+  "regrade": {
+    "utc": "2026-08-18T07:46:49+00:00",
+    "source_report_sha256": "3aab1d1b506b00f8261dda7cfd8f48235608c1dafe74c3e3e60e55424bc97798",
+    "reason": "Compact requested vision answer now accepts the exact '<letter> <height>' form."
+  }
+}