{
  "label": "prefill-safe-int8-mixed-workspace-async-smoke",
  "stamp": "20260615a13",
  "return_codes": {
    "metrics": 0,
    "json_canary": 0,
    "color_canary": 0,
    "quality_suite": 0
  },
  "status": {
    "metrics": "passed",
    "json_canary": "passed",
    "color_canary": "passed",
    "quality_suite": "skipped"
  },
  "artifacts": {
    "metrics": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-smoke-p512o512-20260615a13.json",
    "json_canary": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-smoke-json-repeat32-20260615a13.json",
    "color_canary": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-smoke-color-repeat32-20260615a13.json",
    "quality_suite": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-smoke-quality-suite-20260615a13.json",
    "log": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-smoke-20260615a13.log"
  },
  "artifact_exists": {
    "metrics": true,
    "json_canary": true,
    "color_canary": true,
    "quality_suite": false,
    "log": true
  },
  "env": {
    "VLLM_XPU_GDN_NATIVE_FALLBACK": "prefill",
    "VLLM_XPU_GDN_NATIVE_PREFILL_FALLBACK": null,
    "VLLM_XPU_GDN_PREFILL_RECURRENT_FALLBACK": "1",
    "VLLM_XPU_GDN_NATIVE_FALLBACK_INITIAL_DECODE_STEPS": null,
    "VLLM_XPU_GDN_NATIVE_FALLBACK_DECODE_LAYERS": null,
    "VLLM_XPU_GDN_NATIVE_FALLBACK_DECODE_MAX_LAYER": null,
    "VLLM_XPU_GDN_NATIVE_FALLBACK_DECODE_LAYER_REGEX": null,
    "VLLM_XPU_GDN_SYNC_BEFORE_NATIVE_DECODE": null,
    "VLLM_XPU_GDN_SYNC_BEFORE_NATIVE_DECODE_LAYERS": null,
    "VLLM_XPU_GREEDY_SAMPLE_TOPK_FALLBACK": "1",
    "VLLM_XPU_GREEDY_SAMPLE_XPU_FALLBACK": null,
    "VLLM_XPU_ZERO_ALL_PREFILL_GDN_STATE": null,
    "VLLM_XPU_ZERO_FRESH_GDN_STATE": null,
    "VLLM_XPU_DISABLE_PREFILL_CUDAGRAPH_REPLAY": "1",
    "VLLM_ENABLE_FLA_PACKED_RECURRENT_DECODE": null,
    "XPU_GRAPH": "1",
    "VLLM_XPU_ENABLE_XPU_GRAPH": "1",
    "VLLM_XPU_FORCE_GRAPH_WITH_COMM": "1",
    "VLLM_XPU_GRAPH_NOOP_COMM_CAPTURE": "1",
    "VLLM_XPU_USE_CUSTOM_OP_COLLECTIVES": null,
    "VLLM_XPU_COMPILE_ALLREDUCE_CUSTOM_OP": null,
    "VLLM_XPU_CUSTOM_ALLREDUCE_GRAPH_CLONE_INPUT": null,
    "VLLM_XPU_CUSTOM_ALLREDUCE_CLONE_INPUT": null,
    "VLLM_XPU_CUDAGRAPH_REPLAY_MAX_PIECEWISE_INDEX": null,
    "VLLM_XPU_CUDAGRAPH_DISABLE_SUBMOD_REGEX": null,
    "VLLM_XPU_CUDAGRAPH_RECAPTURE_AFTER_N_REPLAYS": null,
    "VLLM_XPU_CUDAGRAPH_RECAPTURE_REGEX": null,
    "VLLM_XPU_DECODE_CUDAGRAPH_REPLAY_EAGER_EVERY_N_REQUESTS": null,
    "VLLM_XPU_CUDAGRAPH_ALLOW_RUNTIME_RECAPTURE": null,
    "VLLM_XPU_SYNC_CUDAGRAPH_REPLAY": null,
    "VLLM_XPU_CUDAGRAPH_REPLAY_TRACE_FILE": null,
    "VLLM_XPU_CUDAGRAPH_REPLAY_TRACE_MAX_LINES": null,
    "VLLM_XPU_CUDAGRAPH_REPLAY_TRACE_RANK": null,
    "VLLM_XPU_CUDAGRAPH_REPLAY_TRACE_SUBMOD_REGEX": null,
    "VLLM_XPU_CUDAGRAPH_REPLAY_TRACE_REQ_REGEX": null,
    "VLLM_XPU_CUDAGRAPH_REPLAY_TRACE_INPUTS": null,
    "VLLM_XPU_CUDAGRAPH_REPLAY_TRACE_DIGEST": null,
    "VLLM_XPU_GDN_TRACE_FILE": null,
    "VLLM_XPU_GDN_TRACE_MAX_LINES": null,
    "VLLM_XPU_GDN_TRACE_RANK": null,
    "VLLM_XPU_GDN_TRACE_LAYER_REGEX": null,
    "VLLM_XPU_GDN_TRACE_REQ_REGEX": null,
    "VLLM_XPU_GDN_TRACE_PREFILL_ONLY": null,
    "VLLM_XPU_GDN_TRACE_DECODE_ONLY": null,
    "VLLM_XPU_GDN_TRACE_TENSOR_LIMIT": null,
    "VLLM_XPU_GDN_TRACE_STATE_LIMIT": null,
    "VLLM_XPU_CUDAGRAPH_ZERO_REPLAY_OUTPUT_REGEX": null,
    "VLLM_XPU_CUDAGRAPH_ZERO_REPLAY_OUTPUT_INDICES": null,
    "VLLM_XPU_CUDAGRAPH_CLONE_REPLAY_OUTPUT_REGEX": null,
    "VLLM_XPU_CUDAGRAPH_CLONE_REPLAY_OUTPUT_INDICES": null,
    "VLLM_XPU_CUDAGRAPH_COMPARE_DIRECT_REGEX": null,
    "VLLM_XPU_CUDAGRAPH_COMPARE_DIRECT_RETURN_DIRECT": null,
    "VLLM_XPU_W8A8_EXPERIMENTAL_ALLOW": "0",
    "VLLM_XPU_W8A8_USE_OFFSETS": null,
    "VLLM_XPU_W8A8_OFFSETS_PREFIX_OP": null,
    "VLLM_XPU_MOE_W8A8_MIDDLE_LAYERLET": null,
    "VLLM_XPU_INT8_MOE_MIXED_WORKSPACE": "1",
    "GPU_MEMORY_UTILIZATION": "0.90",
    "COMPILATION_CONFIG": "{\"cudagraph_mode\":\"PIECEWISE\"}",
    "ABLATION_FAST_GRAPH_AUTOCONFIG": "1",
    "VLLM_EXTRA_ARGS": "--uvicorn-log-level warning"
  },
  "metrics_summary": {
    "tok_s_out_client_after_first_chunk": {
      "mean": 95.20291502058984,
      "median": 95.20291502058984,
      "min": 95.15760459740709,
      "max": 95.2482254437726
    },
    "tok_s_out_client_after_first_chunk_corrected": {
      "mean": 95.01697182719025,
      "median": 95.01697182719025,
      "min": 94.97174990092778,
      "max": 95.06219375345272
    },
    "tok_s_out_client_e2e": {
      "mean": 92.004380418574,
      "median": 92.004380418574,
      "min": 91.97802608215228,
      "max": 92.03073475499572
    },
    "tok_s_total_client": {
      "mean": 181.49301606007762,
      "median": 181.49301606007762,
      "min": 181.44102801362072,
      "max": 181.54500410653452
    },
    "ttft_ms_client": {
      "mean": 186.96507753338665,
      "median": 186.96507753338665,
      "min": 185.99956505931914,
      "max": 187.93059000745416
    },
    "ttft_ms_vllm_metrics": {
      "mean": 175.11284351348877,
      "median": 175.11284351348877,
      "min": 174.42798614501953,
      "max": 175.797700881958
    },
    "tok_s_prefill_lower_bound_from_ttft": {
      "mean": 2663.670069947083,
      "median": 2663.670069947083,
      "min": 2649.914524188144,
      "max": 2677.4256157060227
    },
    "e2e_ms_vllm_metrics": {
      "mean": 5563.761830329895,
      "median": 5563.761830329895,
      "min": 5561.961889266968,
      "max": 5565.561771392822
    },
    "queue_ms_vllm_histogram": {
      "mean": 0.011230935342609882,
      "median": 0.011230935342609882,
      "min": 0.009697861969470978,
      "max": 0.012764008715748787
    },
    "prefill_ms_vllm_histogram": {
      "mean": 170.09470507036895,
      "median": 170.09470507036895,
      "min": 169.618726009503,
      "max": 170.57068413123488
    },
    "decode_ms_vllm_histogram": {
      "mean": 5388.93856247887,
      "median": 5388.93856247887,
      "min": 5386.47025404498,
      "max": 5391.4068709127605
    },
    "decode_ms_per_generation_token_vllm_histogram": {
      "mean": 10.525270629841543,
      "median": 10.525270629841543,
      "min": 10.520449714931601,
      "max": 10.530091544751485
    },
    "inference_ms_vllm_histogram": {
      "mean": 5559.033267549239,
      "median": 5559.033267549239,
      "min": 5557.040938176215,
      "max": 5561.0255969222635
    },
    "time_per_output_token_ms_vllm_histogram": {
      "mean": 10.545868028334386,
      "median": 10.545868028334386,
      "min": 10.54103767914869,
      "max": 10.550698377520082
    },
    "inter_token_ms_vllm_histogram": {
      "mean": 10.545868028334382,
      "median": 10.545868028334382,
      "min": 10.541037679148689,
      "max": 10.550698377520078
    },
    "iteration_tokens_per_step_vllm_histogram": {
      "mean": 1.97265625,
      "median": 1.97265625,
      "min": 1.97265625,
      "max": 1.97265625
    }
  },
  "speed": {
    "tok_s_out_corrected_mean": 95.01697182719025,
    "tok_s_out_after_first_chunk_mean": 95.20291502058984,
    "tok_s_out_e2e_mean": 92.004380418574,
    "decode_ms_per_token_mean": 10.525270629841543,
    "ttft_ms_client_mean": 186.96507753338665
  },
  "json_canary": {
    "pass_all": true,
    "repeats_completed": 32,
    "mismatch_count": 0,
    "first_mismatch": null
  },
  "color_canary": {
    "pass_all": true,
    "repeats_completed": 32,
    "mismatch_count": 0,
    "first_mismatch": null
  },
  "decision": {
    "accepted_by_requested_gates": true,
    "gate_failures": []
  }
}
