Commit from Azure DevOps update Results

RedHatAI/Qwen3-0.6B-quantized.w4a16-autoround-W4A16/results_2026-04-24-12-25-45.json

@@ -0,0 +1,47 @@
+{
+  "pipeline": "auto_eval",
+  "model_id": "RedHatAI/Qwen3-0.6B-quantized.w4a16",
+  "artifact_name": "Qwen3-0.6B-quantized.w4a16-autoround-W4A16",
+  "generated_at": "2026-04-24T12:25:45Z",
+  "source_runtime_dir": "/root/.openclaw/workspace/quantized/runs/RedHatAI_Qwen3-0.6B-quantized.w4a16-W4A16",
+  "source_model_dir": "/root/_work/1/s/auto_eval/RedHatAI/Qwen3-0.6B-quantized.w4a16",
+  "run_dir": "results/RedHatAI/Qwen3-0.6B-quantized.w4a16-autoround-W4A16/run_2026-04-24-12-25-45",
+  "quant_summary": null,
+  "accuracy": {
+    "model_id": "RedHatAI/Qwen3-0.6B-quantized.w4a16",
+    "model_path": "RedHatAI/Qwen3-0.6B-quantized.w4a16",
+    "scheme": "W4A16",
+    "device": "cuda:0",
+    "num_gpus": "1",
+    "tasks": {
+      "piqa": {
+        "accuracy": 0.6583,
+        "accuracy_stderr": 0.0111
+      },
+      "hellaswag": {
+        "accuracy": 0.3665,
+        "accuracy_stderr": 0.0048
+      },
+      "mmlu": {
+        "accuracy": 0.3375,
+        "accuracy_stderr": 0.0039
+      },
+      "gsm8k": {
+        "accuracy": 0.3798,
+        "accuracy_stderr": 0.0134
+      }
+    },
+    "status": "success",
+    "duration_seconds": 639,
+    "eval_framework": "lm_eval+vllm",
+    "errors": []
+  },
+  "copied_files": [
+    "results/RedHatAI/Qwen3-0.6B-quantized.w4a16-autoround-W4A16/run_2026-04-24-12-25-45/accuracy.json",
+    "results/RedHatAI/Qwen3-0.6B-quantized.w4a16-autoround-W4A16/run_2026-04-24-12-25-45/lm_eval_results",
+    "results/RedHatAI/Qwen3-0.6B-quantized.w4a16-autoround-W4A16/run_2026-04-24-12-25-45/logs",
+    "results/RedHatAI/Qwen3-0.6B-quantized.w4a16-autoround-W4A16/run_2026-04-24-12-25-45/session_eval_327.jsonl",
+    "results/RedHatAI/Qwen3-0.6B-quantized.w4a16-autoround-W4A16/run_2026-04-24-12-25-45/session_eval_327.md"
+  ],
+  "num_gpus": "1"
+}

RedHatAI/Qwen3-0.6B-quantized.w4a16-autoround-W4A16/run_2026-04-24-12-25-45/accuracy.json

@@ -0,0 +1,29 @@
+{
+  "model_id": "RedHatAI/Qwen3-0.6B-quantized.w4a16",
+  "model_path": "RedHatAI/Qwen3-0.6B-quantized.w4a16",
+  "scheme": "W4A16",
+  "device": "cuda:0",
+  "num_gpus": "1",
+  "tasks": {
+    "piqa": {
+      "accuracy": 0.6583,
+      "accuracy_stderr": 0.0111
+    },
+    "hellaswag": {
+      "accuracy": 0.3665,
+      "accuracy_stderr": 0.0048
+    },
+    "mmlu": {
+      "accuracy": 0.3375,
+      "accuracy_stderr": 0.0039
+    },
+    "gsm8k": {
+      "accuracy": 0.3798,
+      "accuracy_stderr": 0.0134
+    }
+  },
+  "status": "success",
+  "duration_seconds": 639,
+  "eval_framework": "lm_eval+vllm",
+  "errors": []
+}

RedHatAI/Qwen3-0.6B-quantized.w4a16-autoround-W4A16/run_2026-04-24-12-25-45/logs/eval_prompt.txt

@@ -0,0 +1,53 @@
+You are an expert in evaluating quantized LLM models.
+You MUST follow the skill instructions in: /root/.openclaw/workspace/skills/auto_eval_vllm/SKILL.md
+
+Quantized model path: RedHatAI/Qwen3-0.6B-quantized.w4a16
+Runtime artifact directory: /root/.openclaw/workspace/quantized/runs/RedHatAI_Qwen3-0.6B-quantized.w4a16-W4A16
+Raw lm_eval output directory: /root/.openclaw/workspace/quantized/runs/RedHatAI_Qwen3-0.6B-quantized.w4a16-W4A16/lm_eval_results
+Evaluation tasks: piqa,mmlu,hellaswag,gsm8k
+Batch size: 8
+Num gpus: 1
+
+The quantized model was produced by auto_quant with scheme=W4A16, export_format=auto_round.
+A venv may already exist at /root/.openclaw/workspace/quantized/runs/RedHatAI_Qwen3-0.6B-quantized.w4a16-W4A16/venv (created by auto_quant with --system-site-packages).
+
+CRITICAL ENVIRONMENT NOTE:
+- System Python has torch+cuda pre-installed. When creating venvs, ALWAYS use:
+    python3 -m venv --system-site-packages <path>
+  This ensures the venv inherits torch+cuda. Do NOT pip install torch inside the venv.
+- If /root/.venv exists, reuse /root/.venv before creating a new venv.
+- If a venv already exists at /root/.openclaw/workspace/quantized/runs/RedHatAI_Qwen3-0.6B-quantized.w4a16-W4A16/venv, reuse it - just install lm_eval and vllm into it.
+- Use uv pip for dependency installation. Prefer:
+    uv pip install --python <venv>/bin/python <packages>
+- Do NOT reinstall torch or flash_attn if they already import successfully from the reused environment. Only install them when missing or incompatible.
+- Write evaluation outputs, logs, prompts, copied request/session files, and other runtime artifacts to: /root/.openclaw/workspace/quantized/runs/RedHatAI_Qwen3-0.6B-quantized.w4a16-W4A16
+- When invoking lm_eval, you MUST pass:
+    --output_path /root/.openclaw/workspace/quantized/runs/RedHatAI_Qwen3-0.6B-quantized.w4a16-W4A16/lm_eval_results
+- Do NOT omit --output_path. Keep the raw lm_eval output files under that exact directory for later upload.
+
+IMPORTANT - After evaluation completes, you MUST produce:
+
+/root/.openclaw/workspace/quantized/runs/RedHatAI_Qwen3-0.6B-quantized.w4a16-W4A16/accuracy.json - evaluation results:
+{
+  "model_id": "RedHatAI/Qwen3-0.6B-quantized.w4a16",
+  "model_path": "RedHatAI/Qwen3-0.6B-quantized.w4a16",
+  "scheme": "W4A16",
+  "device": "cuda:0",
+  "num_gpus": "1",
+  "tasks": {
+    "<task_name>": {
+      "accuracy": <float>,
+      "accuracy_stderr": <float or null>
+    }
+  },
+  "status": "success" or "failed",
+  "duration_seconds": <float>,
+  "eval_framework": "lm_eval+vllm" or "lm_eval+hf" or "manual",
+  "errors": [<list of error strings if any>]
+}
+
+/root/.openclaw/workspace/quantized/runs/RedHatAI_Qwen3-0.6B-quantized.w4a16-W4A16/lm_eval_results/ - raw lm_eval output directory created by:
+    lm_eval ... --output_path /root/.openclaw/workspace/quantized/runs/RedHatAI_Qwen3-0.6B-quantized.w4a16-W4A16/lm_eval_results
+
+The accuracy values MUST be real numbers from actual evaluation runs.
+Write as valid JSON. If evaluation fails, still write accuracy.json with status=failed.