Add run for gemma-4-31B-AutoRound-MXFP8-model_free (gemma-4-31B_MXFP8_20260821-142004_20ac97)

.gitattributes

@@ -203,3 +203,10 @@ results/gemma-4-31B-BF16-Baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/sample
 results/gemma-4-31B-BF16-Baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/samples_mmlu_professional_law_2026-08-21T16-05-46.430867.jsonl filter=lfs diff=lfs merge=lfs -text
 results/gemma-4-31B-BF16-Baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/samples_mmlu_professional_law_2026-08-21T17-39-30.820275.jsonl filter=lfs diff=lfs merge=lfs -text
 results/gemma-4-31B-BF16-Baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/samples_mmlu_professional_law_2026-08-21T19-17-06.515404.jsonl filter=lfs diff=lfs merge=lfs -text
+results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/gsm8k/google__gemma-4-31B/samples_gsm8k_2026-08-21T19-39-04.308398.jsonl filter=lfs diff=lfs merge=lfs -text
+results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/samples_hellaswag_2026-08-21T16-05-46.430867.jsonl filter=lfs diff=lfs merge=lfs -text
+results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/samples_hellaswag_2026-08-21T17-39-30.820275.jsonl filter=lfs diff=lfs merge=lfs -text
+results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/samples_hellaswag_2026-08-21T19-17-06.515404.jsonl filter=lfs diff=lfs merge=lfs -text
+results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/samples_mmlu_professional_law_2026-08-21T16-05-46.430867.jsonl filter=lfs diff=lfs merge=lfs -text
+results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/samples_mmlu_professional_law_2026-08-21T17-39-30.820275.jsonl filter=lfs diff=lfs merge=lfs -text
+results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/samples_mmlu_professional_law_2026-08-21T19-17-06.515404.jsonl filter=lfs diff=lfs merge=lfs -text

knowledge/stats/events.jsonl

@@ -1,3 +1,4 @@
 {"ts": "2026-08-21T06:57:09Z", "phase": "evaluate", "outcome": "agent_fixed", "sig_hash": "", "error_class": "mxfp8_kernel_min_dim_unsupported_layer", "model_id": "Qwen/Qwen3.5-35B-A3B", "tier": "opus"}
 {"ts": "2026-08-21T09:13:09Z", "phase": "evaluate", "outcome": "agent_fixed", "sig_hash": "", "error_class": "mxfp8_kernel_min_dim_unsupported_layer", "model_id": "Qwen/Qwen3.5-9B", "tier": "opus"}
 {"ts": "2026-08-21T12:13:06Z", "phase": "evaluate", "outcome": "agent_fixed", "sig_hash": "", "error_class": "mxfp8_kernel_min_dim_unsupported_layer", "model_id": "Qwen/Qwen3.5-9B", "tier": "opus"}
+{"ts": "2026-08-21T19:39:12Z", "phase": "baseline", "outcome": "agent_fixed", "sig_hash": "", "error_class": "base_model_missing_chat_template", "model_id": "google/gemma-4-31B", "tier": "opus"}

knowledge/trajectories/baseline.jsonl

@@ -0,0 +1 @@
+{"ts": "2026-08-21T19:39:12Z", "phase": "baseline", "signature": "ValueError: Cannot use chat template functions because tokenizer.chat_template is not set and no template argument was passed! For information about w", "error_class": "base_model_missing_chat_template", "exc_type": "ValueError", "outcome": "fixed", "root_cause": "Attempt-4's aqa source edit never executed because the aqa orchestrator", "fix_rationale": "Since the stale harness will keep passing --apply_chat_template, make it SUCCEED:", "component": "environment  (HF model cache: tokenizer_config.json) — force", "smoke_test": "fresh AutoTokenizer.from_pretrained('google/gemma-4-31B') -> chat_template present; apply_chat_template output == lm_eval messages_to_text (byte-exact) for multiturn & singleturn.", "tried_and_rejected": ["heterogeneous_config_per_layer_attr_access: vLLM's Gemma4ModelArchConfigConvertor.get_head_size() reads", "heterogeneous_config_per_layer_attr_access: vLLM's generic config consumers (arch convertor get_* / getattr_iter's", "gemma4_per_layer_head_dim_weight_shape_mismatch: vLLM's gemma4 code reads global attributes global_head_dim /", "base_model_missing_chat_template: aqa's LmEvalHarness.build_command emits --apply_chat_template for every"], "tier": "opus", "steps": 5, "model_id": "google/gemma-4-31B", "scheme": "MXFP8", "method": "model_free", "run_id": "gemma-4-31B_MXFP8_20260821-142004_20ac97"}

results/google/gemma-4-31B-AutoRound-MXFP8-model_free/results_gemma-4-31B_MXFP8_20260821-142004_20ac97.json

@@ -0,0 +1,17 @@
+{
+  "status": "Finished",
+  "pipeline": "aqa",
+  "model": "google/gemma-4-31B",
+  "scheme": "MXFP8",
+  "method": "model_free",
+  "artifact": "gemma-4-31B-AutoRound-MXFP8-model_free",
+  "run_id": "gemma-4-31B_MXFP8_20260821-142004_20ac97",
+  "results": {
+    "piqa": 0.8264417845484222,
+    "hellaswag": 0.8522206731726748,
+    "mmlu": 0.8111380145278451,
+    "gsm8k": 0.8597422289613343
+  },
+  "num_files": 4039,
+  "request_filename": "gemma-4-31B-AutoRound-MXFP8-model_free.json"
+}

results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/.quant_command.json

@@ -0,0 +1 @@
+{"command": "auto-round --model_name google/gemma-4-31B --model_free --scheme MXFP8 --ignore_layers lm_head,embed_tokens,vision_tower,embed_vision,multi_modal_projector --format llm_compressor --output_dir /workspace/aqa/results/gemma-4-31B_MXFP8_20260821-221412_649594/quantized --device_map auto", "source": "agent_optimized"}

results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/aqa_run.json

@@ -0,0 +1 @@
+{"model": "google/gemma-4-31B", "state": "running"}

results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/accuracy.json

@@ -0,0 +1,41 @@
+{
+  "model": "google/gemma-4-31B",
+  "backend": "vllm",
+  "mean_score": 0.8387224564838166,
+  "results": {
+    "piqa": {
+      "score": 0.8291621327529923,
+      "metric": "acc,none",
+      "num_samples": 1838
+    },
+    "hellaswag": {
+      "score": 0.8526190001991635,
+      "metric": "acc_norm,none",
+      "num_samples": 10042
+    },
+    "mmlu": {
+      "score": 0.8118501637943313,
+      "metric": "acc,none",
+      "num_samples": 0
+    },
+    "gsm8k": {
+      "score": 0.8612585291887794,
+      "metric": "exact_match,strict-match",
+      "num_samples": 1319
+    }
+  },
+  "commands": [
+    "lm_eval --model vllm --model_args pretrained=google/gemma-4-31B,tensor_parallel_size=2,max_model_len=8192,gpu_memory_utilization=0.6,dtype=bfloat16,trust_remote_code=True,add_bos_token=True,enable_prefix_caching=False,max_gen_toks=2048,enable_thinking=False --tasks piqa,hellaswag,mmlu --batch_size 1 --output_path /workspace/aqa/results/gemma-4-31B_MXFP8_20260821-221412_649594/baseline/piqa_hellaswag_mmlu --log_samples --seed 42",
+    "lm_eval --model vllm --model_args pretrained=google/gemma-4-31B,tensor_parallel_size=2,max_model_len=8192,gpu_memory_utilization=0.6,dtype=bfloat16,trust_remote_code=True,add_bos_token=True,enable_prefix_caching=False,max_gen_toks=2048,enable_thinking=False --tasks gsm8k --batch_size 64 --output_path /workspace/aqa/results/gemma-4-31B_MXFP8_20260821-221412_649594/baseline/gsm8k --log_samples --seed 42 --num_fewshot 5 --apply_chat_template --fewshot_as_multiturn"
+  ],
+  "healed": [
+    "batch_size=16",
+    "batch_size=8",
+    "batch_size=4",
+    "batch_size=2",
+    "batch_size=1",
+    "gpu_memory_utilization=0.8",
+    "gpu_memory_utilization=0.7",
+    "gpu_memory_utilization=0.6"
+  ]
+}

results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/gsm8k/google__gemma-4-31B/results_2026-08-21T19-39-04.308398.json

@@ -0,0 +1,187 @@
+{
+  "results": {
+    "gsm8k": {
+      "name": "gsm8k",
+      "alias": "gsm8k",
+      "sample_len": 1319,
+      "exact_match,strict-match": 0.8612585291887794,
+      "exact_match_stderr,strict-match": 0.009521649920798143,
+      "exact_match,flexible-extract": 0.8612585291887794,
+      "exact_match_stderr,flexible-extract": 0.009521649920798143
+    }
+  },
+  "group_subtasks": {},
+  "configs": {
+    "gsm8k": {
+      "task": "gsm8k",
+      "dataset_path": "openai/gsm8k",
+      "dataset_name": "main",
+      "training_split": "train",
+      "test_split": "test",
+      "fewshot_split": "train",
+      "doc_to_text": "Question: {{question}}\nAnswer:",
+      "doc_to_target": "{{answer}}",
+      "unsafe_code": false,
+      "description": "",
+      "target_delimiter": " ",
+      "fewshot_delimiter": "\n\n",
+      "fewshot_config": {
+        "sampler": "default",
+        "split": "train",
+        "process_docs": null,
+        "fewshot_indices": null,
+        "samples": null,
+        "doc_to_text": "Question: {{question}}\nAnswer:",
+        "doc_to_choice": null,
+        "doc_to_target": "{{answer}}",
+        "gen_prefix": null,
+        "fewshot_delimiter": "\n\n",
+        "target_delimiter": " "
+      },
+      "num_fewshot": 5,
+      "metric_list": [
+        {
+          "metric": "exact_match",
+          "aggregation": "mean",
+          "higher_is_better": true,
+          "ignore_case": true,
+          "ignore_punctuation": false,
+          "regexes_to_ignore": [
+            ",",
+            "\\$",
+            "(?s).*#### ",
+            "\\.$"
+          ]
+        }
+      ],
+      "output_type": "generate_until",
+      "generation_kwargs": {
+        "until": [
+          "Question:",
+          "</s>",
+          "<|im_end|>"
+        ],
+        "do_sample": false,
+        "temperature": 0.0
+      },
+      "repeats": 1,
+      "filter_list": [
+        {
+          "name": "strict-match",
+          "filter": [
+            {
+              "function": "regex",
+              "regex_pattern": "#### (\\-?[0-9\\.\\,]+)"
+            },
+            {
+              "function": "take_first"
+            }
+          ]
+        },
+        {
+          "name": "flexible-extract",
+          "filter": [
+            {
+              "function": "regex",
+              "group_select": -1,
+              "regex_pattern": "(-?[$0-9.,]{2,})|(-?[0-9]+)"
+            },
+            {
+              "function": "take_first"
+            }
+          ]
+        }
+      ],
+      "should_decontaminate": false,
+      "metadata": {
+        "version": 3.0,
+        "pretrained": "google/gemma-4-31B",
+        "tensor_parallel_size": 2,
+        "max_model_len": 8192,
+        "gpu_memory_utilization": 0.6,
+        "dtype": "bfloat16",
+        "trust_remote_code": true,
+        "add_bos_token": true,
+        "enable_prefix_caching": false,
+        "max_gen_toks": 2048,
+        "enable_thinking": false,
+        "config_source": "/usr/local/lib/python3.12/dist-packages/lm_eval/tasks/gsm8k/gsm8k.yaml"
+      }
+    }
+  },
+  "versions": {
+    "gsm8k": 3.0
+  },
+  "n-shot": {
+    "gsm8k": 5
+  },
+  "higher_is_better": {
+    "gsm8k": {
+      "exact_match": true
+    }
+  },
+  "n-samples": {
+    "gsm8k": {
+      "original": 1319,
+      "effective": 1319
+    }
+  },
+  "config": {
+    "model": "vllm",
+    "model_args": {
+      "pretrained": "google/gemma-4-31B",
+      "tensor_parallel_size": 2,
+      "max_model_len": 8192,
+      "gpu_memory_utilization": 0.6,
+      "dtype": "bfloat16",
+      "trust_remote_code": true,
+      "add_bos_token": true,
+      "enable_prefix_caching": false,
+      "max_gen_toks": 2048,
+      "enable_thinking": false
+    },
+    "batch_size": "64",
+    "batch_sizes": [],
+    "device": "cuda:0",
+    "use_cache": null,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "gen_kwargs": {},
+    "random_seed": 42,
+    "numpy_seed": 42,
+    "torch_seed": 42,
+    "fewshot_seed": 42
+  },
+  "git_hash": null,
+  "date": 1787339835.801998,
+  "pretty_env_info": "PyTorch version: 2.11.0+cu130\nIs debug build: False\nCUDA used to build PyTorch: 13.0\nROCM used to build PyTorch: N/A\n\nOS: Ubuntu 24.04.3 LTS (x86_64)\nGCC version: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0\nClang version: Could not collect\nCMake version: version 3.31.6\nLibc version: glibc-2.39\n\nPython version: 3.12.3 (main, Nov  6 2025, 13:44:16) [GCC 13.3.0] (64-bit runtime)\nPython platform: Linux-6.8.0-136-generic-x86_64-with-glibc2.39\nIs CUDA available: True\nCUDA runtime version: 13.1.115\nCUDA_MODULE_LOADING set to: LAZY\nGPU models and configuration: \nGPU 0: NVIDIA RTX PRO 5000 72GB Blackwell\nGPU 1: NVIDIA RTX PRO 5000 72GB Blackwell\n\nNvidia driver version: Could not collect\ncuDNN version: Probably one of the following:\n/usr/lib/x86_64-linux-gnu/libcudnn.so.9.17.1\n/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.17.1\n/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.17.1\n/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.17.1\n/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.17.1\n/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.17.1\n/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.17.1\n/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.17.1\nIs XPU available: False\nHIP runtime version: N/A\nMIOpen runtime version: N/A\nIs XNNPACK available: True\nCaching allocator config: N/A\n\nCPU:\nArchitecture:                            x86_64\nCPU op-mode(s):                          32-bit, 64-bit\nAddress sizes:                           52 bits physical, 57 bits virtual\nByte Order:                              Little Endian\nCPU(s):                                  256\nOn-line CPU(s) list:                     0-255\nVendor ID:                               GenuineIntel\nModel name:                              Intel(R) Xeon(R) 6767P\nCPU family:                              6\nModel:                                   173\nThread(s) per core:                      2\nCore(s) per socket:                      64\nSocket(s):                               2\nStepping:                                1\nCPU(s) scaling MHz:                      100%\nCPU max MHz:                             3900.0000\nCPU min MHz:                             800.0000\nBogoMIPS:                                4800.00\nFlags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cat_l2 cdp_l3 intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect user_shstk avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req hfi vnmi avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr ibt amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities ibpb_exit_to_user\nVirtualization:                          VT-x\nL1d cache:                               6 MiB (128 instances)\nL1i cache:                               8 MiB (128 instances)\nL2 cache:                                256 MiB (128 instances)\nL3 cache:                                672 MiB (2 instances)\nNUMA node(s):                            2\nNUMA node0 CPU(s):                       0-63,128-191\nNUMA node1 CPU(s):                       64-127,192-255\nVulnerability Gather data sampling:      Not affected\nVulnerability Indirect target selection: Not affected\nVulnerability Itlb multihit:             Not affected\nVulnerability L1tf:                      Not affected\nVulnerability Mds:                       Not affected\nVulnerability Meltdown:                  Not affected\nVulnerability Mmio stale data:           Not affected\nVulnerability Reg file data sampling:    Not affected\nVulnerability Retbleed:                  Not affected\nVulnerability Spec rstack overflow:      Not affected\nVulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl\nVulnerability Spectre v1:                Mitigation; usercopy/swapgs barriers and __user pointer sanitization\nVulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS Not affected; BHI BHI_DIS_S\nVulnerability Srbds:                     Not affected\nVulnerability Tsa:                       Not affected\nVulnerability Tsx async abort:           Not affected\nVulnerability Vmscape:                   Mitigation; IBPB before exit to userspace\n\nVersions of relevant libraries:\n[pip3] intel-openmp==2021.4.0\n[pip3] mkl==2021.1.1\n[pip3] mkl-devel==2021.1.1\n[pip3] mkl-include==2021.1.1\n[pip3] mypy_extensions==1.1.0\n[pip3] numpy==2.1.0\n[pip3] nvidia-cublas==13.1.0.3\n[pip3] nvidia-cuda-cupti==13.0.85\n[pip3] nvidia-cuda-nvrtc==13.0.88\n[pip3] nvidia-cuda-runtime==13.0.96\n[pip3] nvidia-cuda-runtime-cu13==0.0.0a0\n[pip3] nvidia-cudnn-cu13==9.19.0.56\n[pip3] nvidia-cudnn-frontend==1.17.0\n[pip3] nvidia-cufft==12.0.0.61\n[pip3] nvidia-curand==10.4.0.35\n[pip3] nvidia-cusolver==12.0.4.66\n[pip3] nvidia-cusparse==12.6.3.3\n[pip3] nvidia-cusparselt-cu13==0.8.0\n[pip3] nvidia-nccl-cu13==2.28.9\n[pip3] nvidia-nvjitlink==13.0.88\n[pip3] nvidia-nvtx==13.0.85\n[pip3] nvtx==0.2.14\n[pip3] onnx==1.18.0\n[pip3] onnx-ir==0.1.14\n[pip3] onnxscript==0.5.7\n[pip3] optree==0.18.0\n[pip3] pytorch-triton==3.6.0+git5261b273.nv26.1\n[pip3] tbb==2021.13.1\n[pip3] tokenspeed-triton==3.8.10.post20260721\n[pip3] torch==2.11.0\n[pip3] torch_c_dlpack_ext==0.1.5\n[pip3] torch_tensorrt==2.10.0a0\n[pip3] torchao==0.15.0+git1272f3cf\n[pip3] torchaudio==2.11.0\n[pip3] torchdata==0.11.0\n[pip3] torchprofile==0.0.4\n[pip3] torchtitan==0.2.0+gite98ae995\n[pip3] torchvision==0.26.0\n[pip3] triton==3.6.0\n[pip3] triton_kernels==1.0.0+git5261b273.nv26.1\n[conda] Could not collect",
+  "transformers_version": "5.15.1",
+  "lm_eval_version": "0.4.12",
+  "upper_git_hash": null,
+  "tokenizer_pad_token": [
+    "<pad>",
+    "0"
+  ],
+  "tokenizer_eos_token": [
+    "<eos>",
+    "1"
+  ],
+  "tokenizer_bos_token": [
+    "<bos>",
+    "2"
+  ],
+  "eot_token_id": 1,
+  "max_length": 8192,
+  "task_hashes": {
+    "gsm8k": "547e5ab01e44b306c8d1f1ad5cbafeaa2b9e48d00a75211e8a03cd19c33799aa"
+  },
+  "model_source": "vllm",
+  "model_name": "google/gemma-4-31B",
+  "model_name_sanitized": "google__gemma-4-31B",
+  "system_instruction": null,
+  "system_instruction_sha": null,
+  "fewshot_as_multiturn": true,
+  "chat_template": "{%- for message in messages -%}{%- if message['role'] == 'assistant' -%}{{ ' ' + message['content'] + '\\n\\n' }}{%- elif message['role'] == 'system' -%}{{ message['content'] + '\\n\\n' }}{%- else -%}{{ message['content'] }}{%- endif -%}{%- endfor -%}",
+  "chat_template_sha": "a66498b07e5f3962e95240bb5e4dabae42451f574d49818af383a193ed7b360b",
+  "total_evaluation_time_seconds": "1312.92314497201"
+}

results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/gsm8k/google__gemma-4-31B/samples_gsm8k_2026-08-21T19-39-04.308398.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8a005336881b95db4b936fe5052c2a7d5fd3620783978561ad228143a73b689
+size 12460981

results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/gsm8k/gsm8k.eval.log

@@ -0,0 +1,141 @@
+/usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.6.3) or chardet (6.0.0.post1)/charset_normalizer (3.4.4) doesn't match a supported version!
+  warnings.warn(
+2026-08-21:19:17:15 INFO     [_cli.run:388] Selected Tasks: ['gsm8k']
+2026-08-21:19:17:15 INFO     [evaluator:214] Setting random seed to 42 | Setting numpy seed to 42 | Setting torch manual seed to 42 | Setting fewshot manual seed to 42
+2026-08-21:19:17:15 INFO     [evaluator:239] Initializing vllm model, with arguments: {'pretrained': 'google/gemma-4-31B', 'tensor_parallel_size': 2, 'max_model_len': 8192, 'gpu_memory_utilization': 0.6, 'dtype': 'bfloat16', 'trust_remote_code': True, 'add_bos_token': True, 'enable_prefix_caching': False, 'max_gen_toks': 2048, 'enable_thinking': False}
+INFO 08-21 19:17:21 [utils.py:278] non-default args: {'trust_remote_code': True, 'dtype': 'bfloat16', 'seed': 1234, 'max_model_len': 8192, 'tensor_parallel_size': 2, 'enable_prefix_caching': False, 'gpu_memory_utilization': 0.6, 'disable_log_stats': True, 'model': 'google/gemma-4-31B'}
+[transformers] Reading global config value for per-layer attribute `num_key_value_heads` on a heterogeneous config. Only do this if the caller can safely handle heterogeneous configs; code that assumes a homogeneous model may use the global value incorrectly.
+INFO 08-21 19:17:27 [model.py:617] Resolved architecture: Gemma4ForConditionalGeneration
+INFO 08-21 19:17:27 [model.py:1752] Using max model len 8192
+INFO 08-21 19:17:33 [scheduler.py:239] Chunked prefill is enabled with max_num_batched_tokens=16384.
+[transformers] Reading global config value for per-layer attribute `head_dim` on a heterogeneous config. Only do this if the caller can safely handle heterogeneous configs; code that assumes a homogeneous model may use the global value incorrectly.
+INFO 08-21 19:17:33 [vllm.py:977] Asynchronous scheduling is enabled.
+INFO 08-21 19:17:33 [kernel.py:270] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'])
+WARNING 08-21 19:17:33 [cuda.py:243] Forcing --disable_chunked_mm_input for models with multimodal-bidirectional attention.
+/usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.6.3) or chardet (6.0.0.post1)/charset_normalizer (3.4.4) doesn't match a supported version!
+  warnings.warn(
+(EngineCore pid=13746) INFO 08-21 19:17:45 [core.py:112] Initializing a V1 LLM engine (v0.22.1) with config: model='google/gemma-4-31B', speculative_config=None, tokenizer='google/gemma-4-31B', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=True, dtype=torch.bfloat16, max_seq_len=8192, download_dir=None, load_format=auto, tensor_parallel_size=2, pipeline_parallel_size=1, data_parallel_size=1, decode_context_parallel_size=1, dcp_comm_backend=ag_rs, disable_custom_all_reduce=False, quantization=None, quantization_config=None, enforce_eager=False, enable_return_routed_experts=False, kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, kv_cache_metrics=False, kv_cache_metrics_sample=0.01, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False), seed=1234, served_model_name=google/gemma-4-31B, enable_prefix_caching=False, enable_chunked_prefill=True, pooler_config=None, compilation_config={'mode': <CompilationMode.VLLM_COMPILE: 3>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['none'], 'ir_enable_torch_wrap': True, 'splitting_ops': ['vllm::unified_attention_with_output', 'vllm::unified_mla_attention_with_output', 'vllm::mamba_mixer2', 'vllm::mamba_mixer', 'vllm::short_conv', 'vllm::linear_attention', 'vllm::plamo2_mamba_mixer', 'vllm::qwen_gdn_attention_core', 'vllm::gdn_attention_core_xpu', 'vllm::olmo_hybrid_gdn_full_forward', 'vllm::kda_attention', 'vllm::sparse_attn_indexer', 'vllm::rocm_aiter_sparse_attn_indexer', 'vllm::deepseek_v4_attention', 'vllm::unified_kv_cache_update', 'vllm::unified_mla_kv_cache_update'], 'compile_mm_encoder': False, 'cudagraph_mm_encoder': False, 'encoder_cudagraph_token_budgets': [], 'encoder_cudagraph_max_vision_items_per_batch': 0, 'encoder_cudagraph_max_frames_per_batch': None, 'compile_sizes': [], 'compile_ranges_endpoints': [16384], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'size_asserts': False, 'alignment_asserts': False, 'scalar_asserts': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.FULL_AND_PIECEWISE: (2, 1)>, 'cudagraph_num_of_warmups': 1, 'cudagraph_capture_sizes': [1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136, 144, 152, 160, 168, 176, 184, 192, 200, 208, 216, 224, 232, 240, 248, 256, 272, 288, 304, 320, 336, 352, 368, 384, 400, 416, 432, 448, 464, 480, 496, 512], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': False, 'fuse_act_quant': False, 'fuse_attn_quant': False, 'enable_sp': False, 'fuse_gemm_comms': False, 'fuse_allreduce_rms': False, 'fuse_rope_kvcache_cat_mla': False, 'fuse_act_padding': False}, 'max_cudagraph_capture_size': 512, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False, 'assume_32_bit_indexing': False}, 'local_cache_dir': None, 'fast_moe_cold_start': False, 'static_all_moe_layers': []}, kernel_config=KernelConfig(ir_op_priority=IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native']), enable_flashinfer_autotune=True, moe_backend='auto', linear_backend='auto')
+(EngineCore pid=13746) WARNING 08-21 19:17:45 [multiproc_executor.py:1029] Reducing Torch parallelism from 128 threads to 1 to avoid unnecessary CPU contention. Set OMP_NUM_THREADS in the external environment to tune this value as needed.
+(EngineCore pid=13746) INFO 08-21 19:17:45 [multiproc_executor.py:139] DP group leader: node_rank=0, node_rank_within_dp=0, master_addr=127.0.0.1, mq_connect_ip=172.17.0.2 (local), world_size=2, local_world_size=2
+/usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.6.3) or chardet (6.0.0.post1)/charset_normalizer (3.4.4) doesn't match a supported version!
+  warnings.warn(
+/usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.6.3) or chardet (6.0.0.post1)/charset_normalizer (3.4.4) doesn't match a supported version!
+  warnings.warn(
+(Worker pid=14011) INFO 08-21 19:18:01 [parallel_state.py:1422] world_size=2 rank=1 local_rank=1 distributed_init_method=tcp://127.0.0.1:35421 backend=nccl
+(Worker pid=14010) INFO 08-21 19:18:02 [parallel_state.py:1422] world_size=2 rank=0 local_rank=0 distributed_init_method=tcp://127.0.0.1:35421 backend=nccl
+(Worker pid=14010) INFO 08-21 19:18:04 [pynccl.py:113] vLLM is using nccl==2.28.9
+(Worker pid=14010) WARNING 08-21 19:18:04 [symm_mem.py:66] SymmMemCommunicator: Device capability 12.0 not supported, communicator is not available.
+(Worker pid=14011) WARNING 08-21 19:18:04 [symm_mem.py:66] SymmMemCommunicator: Device capability 12.0 not supported, communicator is not available.
+(Worker pid=14010) INFO 08-21 19:18:04 [cuda_communicator.py:232] Using ['CUSTOM', 'PYNCCL'] all-reduce backends (in dispatch order) for group 'tp:0' out of potential backends: ['NCCL_SYMM_MEM', 'QUICK_REDUCE', 'FLASHINFER', 'CUSTOM', 'SYMM_MEM', 'PYNCCL'].
+(Worker pid=14010) INFO 08-21 19:18:04 [parallel_state.py:1735] rank 0 in world size 2 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank N/A, EPLB rank N/A
+(Worker pid=14010) INFO 08-21 19:18:04 [topk_topp_sampler.py:45] Using FlashInfer for top-p & top-k sampling.
+(Worker_TP0 pid=14010) INFO 08-21 19:18:05 [gpu_model_runner.py:5037] Starting to load model google/gemma-4-31B...
+(Worker_TP0 pid=14010) [transformers] Reading global config value for per-layer attribute `num_key_value_heads` on a heterogeneous config. Only do this if the caller can safely handle heterogeneous configs; code that assumes a homogeneous model may use the global value incorrectly.
+(Worker_TP0 pid=14010) INFO 08-21 19:18:05 [vllm.py:977] Asynchronous scheduling is enabled.
+(Worker_TP0 pid=14010) INFO 08-21 19:18:05 [kernel.py:270] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'])
+(Worker_TP0 pid=14010) [transformers] Reading global config value for per-layer attribute `head_dim` on a heterogeneous config. Only do this if the caller can safely handle heterogeneous configs; code that assumes a homogeneous model may use the global value incorrectly.
+(Worker_TP1 pid=14011) [transformers] Reading global config value for per-layer attribute `num_key_value_heads` on a heterogeneous config. Only do this if the caller can safely handle heterogeneous configs; code that assumes a homogeneous model may use the global value incorrectly.
+(Worker_TP1 pid=14011) INFO 08-21 19:18:05 [kernel.py:270] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'])
+(Worker_TP1 pid=14011) [transformers] Reading global config value for per-layer attribute `head_dim` on a heterogeneous config. Only do this if the caller can safely handle heterogeneous configs; code that assumes a homogeneous model may use the global value incorrectly.
+(Worker_TP0 pid=14010) INFO 08-21 19:18:05 [cuda.py:378] Using TRITON_ATTN attention backend out of potential backends: ['TRITON_ATTN', 'FLEX_ATTENTION'].
+(Worker_TP0 pid=14010) INFO 08-21 19:18:09 [weight_utils.py:922] Filesystem type for checkpoints: EXT4. Checkpoint size: 58.25 GiB. Available RAM: 438.84 GiB.
+(Worker_TP0 pid=14010) INFO 08-21 19:18:09 [weight_utils.py:945] Auto-prefetch is disabled because the filesystem (EXT4) is not a recognized network FS (NFS/Lustre). If you want to force prefetching, start vLLM with --safetensors-load-strategy=prefetch.
+(Worker_TP0 pid=14010) 
+Loading safetensors checkpoint shards:   0% 0/2 [00:00<?, ?it/s]
+Loading safetensors checkpoint shards: 100% 2/2 [00:09<00:00,  4.86s/it]
+(Worker_TP0 pid=14010) INFO 08-21 19:18:19 [default_loader.py:397] Loading weights took 9.93 seconds
+(Worker_TP0 pid=14010) INFO 08-21 19:18:19 [gpu_model_runner.py:5132] Model loading took 30.38 GiB memory and 13.969650 seconds
+(Worker_TP0 pid=14010) INFO 08-21 19:18:20 [gpu_model_runner.py:6136] Encoder cache will be initialized with a budget of 16384 tokens, and profiled with 6 video items of the maximum feature size.
+(Worker_TP0 pid=14010) INFO 08-21 19:19:19 [backends.py:1089] Using cache directory: /root/.cache/vllm/torch_compile_cache/2818d9d014/rank_0_0/backbone for vLLM's torch.compile
+(Worker_TP0 pid=14010) INFO 08-21 19:19:19 [backends.py:1148] Dynamo bytecode transform time: 3.08 s
+(EngineCore pid=13746) INFO 08-21 19:19:20 [shm_broadcast.py:698] No available shared memory broadcast block found in 60 seconds. This typically happens when some processes are hanging or doing some time-consuming work (e.g. compilation, weight/kv cache quantization).
+(Worker_TP0 pid=14010) INFO 08-21 19:19:21 [backends.py:292] Directly load the compiled graph(s) for compile range (1, 16384) from the cache, took 1.724 s
+(Worker_TP0 pid=14010) INFO 08-21 19:19:21 [decorators.py:311] Directly load AOT compilation from path /root/.cache/vllm/torch_compile_cache/torch_aot_compile/6610a130de4c01c69a5c84e804b0f524e618cf5ca86139442a468080330116c2/rank_0_0/model
+(Worker_TP0 pid=14010) INFO 08-21 19:19:21 [monitor.py:53] torch.compile took 5.27 s in total
+(Worker_TP0 pid=14010) INFO 08-21 19:19:21 [monitor.py:81] Initial profiling/warmup run took 0.02 s
+(Worker_TP1 pid=14011) INFO 08-21 19:19:26 [decorators.py:311] Directly load AOT compilation from path /root/.cache/vllm/torch_compile_cache/torch_aot_compile/6610a130de4c01c69a5c84e804b0f524e618cf5ca86139442a468080330116c2/rank_1_0/model
+(Worker_TP1 pid=14011) INFO 08-21 19:19:30 [gpu_model_runner.py:6279] Profiling CUDA graph memory: PIECEWISE=51 (largest=512), FULL=51 (largest=512)
+(Worker_TP0 pid=14010) INFO 08-21 19:19:30 [gpu_model_runner.py:6279] Profiling CUDA graph memory: PIECEWISE=51 (largest=512), FULL=51 (largest=512)
+(Worker_TP1 pid=14011) INFO 08-21 19:19:30 [custom_all_reduce.py:215] Registering 480 cuda graph addresses
+(Worker_TP0 pid=14010) INFO 08-21 19:19:31 [custom_all_reduce.py:215] Registering 480 cuda graph addresses
+(Worker_TP0 pid=14010) INFO 08-21 19:19:31 [gpu_model_runner.py:6365] Estimated CUDA graph memory: 1.76 GiB total
+(Worker_TP1 pid=14011) INFO 08-21 19:19:31 [gpu_model_runner.py:6365] Estimated CUDA graph memory: 1.76 GiB total
+(Worker_TP0 pid=14010) INFO 08-21 19:19:31 [gpu_worker.py:466] Available KV cache memory: 6.28 GiB
+(Worker_TP0 pid=14010) INFO 08-21 19:19:31 [gpu_worker.py:481] CUDA graph memory profiling is enabled (default since v0.21.0). The current --gpu-memory-utilization=0.6000 is equivalent to --gpu-memory-utilization=0.5753 without CUDA graph memory profiling. To maintain the same effective KV cache size as before, increase --gpu-memory-utilization to 0.6247. To disable, set VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0.
+(Worker_TP1 pid=14011) INFO 08-21 19:19:32 [gpu_worker.py:481] CUDA graph memory profiling is enabled (default since v0.21.0). The current --gpu-memory-utilization=0.6000 is equivalent to --gpu-memory-utilization=0.5753 without CUDA graph memory profiling. To maintain the same effective KV cache size as before, increase --gpu-memory-utilization to 0.6247. To disable, set VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0.
+(EngineCore pid=13746) INFO 08-21 19:19:32 [kv_cache_utils.py:1733] GPU KV cache size: 14,946 tokens
+(EngineCore pid=13746) INFO 08-21 19:19:32 [kv_cache_utils.py:1734] Maximum concurrency for 8,192 tokens per request: 1.82x
+(Worker_TP0 pid=14010) 2026-08-21 19:19:32,024 - INFO - autotuner.py:615 - flashinfer.jit: [Autotuner]: Autotuning process starts ...
+(Worker_TP1 pid=14011) 2026-08-21 19:19:32,024 - INFO - autotuner.py:615 - flashinfer.jit: [Autotuner]: Autotuning process starts ...
+(Worker_TP1 pid=14011) 2026-08-21 19:19:32,046 - INFO - autotuner.py:634 - flashinfer.jit: [Autotuner]: Autotuning process ends
+(Worker_TP0 pid=14010) 2026-08-21 19:19:32,046 - INFO - autotuner.py:634 - flashinfer.jit: [Autotuner]: Autotuning process ends
+(Worker_TP0 pid=14010) 
+Capturing CUDA graphs (mixed prefill-decode, PIECEWISE):   0% 0/51 [00:00<?, ?it/s]
+Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100% 51/51 [00:04<00:00, 11.16it/s]
+(Worker_TP0 pid=14010) 
+Capturing CUDA graphs (decode, FULL):   0% 0/51 [00:00<?, ?it/s]
+Capturing CUDA graphs (decode, FULL): 100% 51/51 [00:03<00:00, 13.77it/s]
+(Worker_TP0 pid=14010) INFO 08-21 19:19:43 [custom_all_reduce.py:215] Registering 12240 cuda graph addresses
+(Worker_TP1 pid=14011) INFO 08-21 19:19:43 [custom_all_reduce.py:215] Registering 12240 cuda graph addresses
+(Worker_TP1 pid=14011) INFO 08-21 19:19:43 [gpu_worker.py:619] CUDA graph pool memory: 0.61 GiB (actual), 1.76 GiB (estimated), difference: 1.14 GiB (186.3%).
+(Worker_TP1 pid=14011) INFO 08-21 19:19:43 [jit_monitor.py:54] Kernel JIT monitor activated — Triton JIT compilations during inference will be logged as warnings.
+(Worker_TP0 pid=14010) INFO 08-21 19:19:43 [gpu_model_runner.py:6456] Graph capturing finished in 12 secs, took 0.61 GiB
+(Worker_TP0 pid=14010) INFO 08-21 19:19:43 [gpu_worker.py:619] CUDA graph pool memory: 0.61 GiB (actual), 1.76 GiB (estimated), difference: 1.14 GiB (186.3%).
+(Worker_TP0 pid=14010) INFO 08-21 19:19:43 [jit_monitor.py:54] Kernel JIT monitor activated — Triton JIT compilations during inference will be logged as warnings.
+(EngineCore pid=13746) INFO 08-21 19:19:43 [core.py:302] init engine (profile, create kv cache, warmup model) took 83.99 s (compilation: 5.27 s)
+(EngineCore pid=13746) INFO 08-21 19:19:51 [vllm.py:977] Asynchronous scheduling is enabled.
+(EngineCore pid=13746) INFO 08-21 19:19:51 [kernel.py:270] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'])
+2026-08-21:19:20:23 INFO     [evaluator_utils:446] Selected tasks:
+2026-08-21:19:20:23 INFO     [evaluator_utils:480] Task: gsm8k (gsm8k/gsm8k.yaml)
+2026-08-21:19:20:23 INFO     [evaluator:314] gsm8k: Using gen_kwargs: {'until': ['Question:', '</s>', '<|im_end|>'], 'do_sample': False, 'temperature': 0.0}
+2026-08-21:19:20:23 WARNING  [evaluator:333] Overwriting default num_fewshot of gsm8k from 5 to 5
+2026-08-21:19:20:23 INFO     [api.task:312] Building contexts for gsm8k on rank 0...
+
+  0% 0/1319 [00:00<?, ?it/s]
+100% 1319/1319 [00:03<00:00, 391.11it/s]
+2026-08-21:19:20:26 INFO     [evaluator:585] Running generate_until requests
+
+Running generate_until requests:   0% 0/1319 [00:00<?, ?it/s](Worker_TP0 pid=14010) WARNING 08-21 19:20:27 [jit_monitor.py:103] Triton kernel JIT compilation during inference: _compute_slot_mapping_kernel. This causes a latency spike; consider extending warmup to cover this shape/config.
+(Worker_TP0 pid=14010) WARNING 08-21 19:20:27 [jit_monitor.py:103] Triton kernel JIT compilation during inference: kernel_unified_attention. This causes a latency spike; consider extending warmup to cover this shape/config.
+
+Running generate_until requests:   5% 65/1319 [02:27<47:18,  2.26s/it]
+Running generate_until requests:  10% 128/1319 [02:44<44:55,  2.26s/it]
+Running generate_until requests:  15% 193/1319 [04:48<26:25,  1.41s/it]
+Running generate_until requests:  19% 256/1319 [05:04<24:56,  1.41s/it]
+Running generate_until requests:  24% 321/1319 [06:50<19:34,  1.18s/it]
+Running generate_until requests:  29% 384/1319 [07:04<18:20,  1.18s/it]
+Running generate_until requests:  39% 513/1319 [10:13<14:57,  1.11s/it]
+Running generate_until requests:  44% 576/1319 [10:24<13:47,  1.11s/it]
+Running generate_until requests:  49% 641/1319 [12:20<12:07,  1.07s/it]
+Running generate_until requests:  53% 704/1319 [12:34<10:59,  1.07s/it]
+Running generate_until requests:  63% 833/1319 [14:46<07:37,  1.06it/s]
+Running generate_until requests:  68% 896/1319 [15:04<06:38,  1.06it/s]
+Running generate_until requests:  83% 1089/1319 [17:04<02:55,  1.31it/s]
+Running generate_until requests:  87% 1152/1319 [17:14<02:07,  1.31it/s]
+Running generate_until requests: 100% 1319/1319 [18:29<00:00,  1.19it/s]
+fatal: not a git repository (or any parent up to mount point /workspace)
+Stopping at filesystem boundary (GIT_DISCOVERY_ACROSS_FILESYSTEM not set).
+(EngineCore pid=13746) INFO 08-21 19:38:59 [core.py:1266] Shutdown initiated (timeout=0)
+(EngineCore pid=13746) INFO 08-21 19:38:59 [core.py:1289] Shutdown complete
+(Worker_TP0 pid=14010) INFO 08-21 19:38:59 [multiproc_executor.py:775] Parent process exited, terminating worker queues
+(Worker_TP1 pid=14011) INFO 08-21 19:38:59 [multiproc_executor.py:872] WorkerProc shutting down.
+(Worker_TP0 pid=14010) INFO 08-21 19:38:59 [multiproc_executor.py:872] WorkerProc shutting down.
+[rank1]:[W821 19:39:00.630710424 TCPStore.cpp:125] [c10d] recvValue failed on SocketImpl(fd=56, addr=[localhost]:54476, remote=[localhost]:35421): Failed to recv, got 0 bytes. Connection was likely closed. Did the remote server shutdown or crash?
+Exception raised from recvBytes at /pytorch/torch/csrc/distributed/c10d/Utils.hpp:682 (most recent call first):
+frame #0: c10::Error::Error(c10::SourceLocation, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >) + 0x9d (0x7c2ec3f7305d in /usr/local/lib/python3.12/dist-packages/torch/lib/libc10.so)
+frame #1: <unknown function> + 0x6a914bd (0x7c2e2ccb24bd in /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_cpu.so)
+frame #2: c10d::TCPStore::check(std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > > const&) + 0x273 (0x7c2e2ccb0413 in /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_cpu.so)
+frame #3: c10d::ProcessGroupNCCL::HeartbeatMonitor::runLoop() + 0x4a5 (0x7c2e0f49c8c5 in /usr/local/lib/python3.12/dist-packages/torch/lib/libtorch_cuda.so)
+frame #4: <unknown function> + 0xecdb4 (0x7c2ec56c2db4 in /lib/x86_64-linux-gnu/libstdc++.so.6)
+frame #5: <unknown function> + 0x9caa4 (0x7c2ec863eaa4 in /lib/x86_64-linux-gnu/libc.so.6)
+frame #6: __clone + 0x44 (0x7c2ec86cba64 in /lib/x86_64-linux-gnu/libc.so.6)
+
+[rank1]:[W821 19:39:00.632668463 ProcessGroupNCCL.cpp:1826] [PG ID 0 PG GUID 0 Rank 1] Failed to check the "should dump" flag on TCPStore, (maybe TCPStore server has shut down too early), with error: Failed to recv, got 0 bytes. Connection was likely closed. Did the remote server shutdown or crash?
+2026-08-21:19:39:04 INFO     [loggers.evaluation_tracker:247] Saving results aggregated
+2026-08-21:19:39:04 INFO     [loggers.evaluation_tracker:119] Saving per-task samples to /workspace/aqa/results/gemma-4-31B_MXFP8_20260821-221412_649594/baseline/gsm8k/google__gemma-4-31B/*.jsonl
+vllm ({'pretrained': 'google/gemma-4-31B', 'tensor_parallel_size': 2, 'max_model_len': 8192, 'gpu_memory_utilization': 0.6, 'dtype': 'bfloat16', 'add_bos_token': True, 'enable_prefix_caching': False, 'max_gen_toks': 2048, 'enable_thinking': False}), gen_kwargs: ({}), limit: None, num_fewshot: 5, batch_size: 64
+|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value |   |Stderr|
+|-----|------:|----------------|-----:|-----------|---|-----:|---|-----:|
+|gsm8k|      3|flexible-extract|     5|exact_match|↑  |0.8613|±  |0.0095|
+|     |       |strict-match    |     5|exact_match|↑  |0.8613|±  |0.0095|
+

results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/samples_hellaswag_2026-08-21T16-05-46.430867.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dba60b2434a0727e7d23f34e612cdc953d35d59e17d0e575e7fb89d845973ce2
+size 43664042

results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/samples_hellaswag_2026-08-21T17-39-30.820275.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f1f8311fbb68e55cf65bc2d77569a73c32bf812e6da62369f53535d063b1187
+size 43664282

results/google/gemma-4-31B-AutoRound-MXFP8-model_free/run_gemma-4-31B_MXFP8_20260821-142004_20ac97/baseline/piqa_hellaswag_mmlu/google__gemma-4-31B/samples_hellaswag_2026-08-21T19-17-06.515404.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f1f8311fbb68e55cf65bc2d77569a73c32bf812e6da62369f53535d063b1187
+size 43664282