{
  "label": "prefill-safe-int8-mixed-workspace-async-deep-gate",
  "stamp": "20260615a13deep2",
  "return_codes": {
    "metrics": 0,
    "json_canary": 0,
    "color_canary": 0,
    "quality_suite": 0
  },
  "status": {
    "metrics": "passed",
    "json_canary": "passed",
    "color_canary": "passed",
    "quality_suite": "passed"
  },
  "artifacts": {
    "metrics": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-deep-gate-p512o512-20260615a13deep2.json",
    "json_canary": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-deep-gate-json-repeat128-20260615a13deep2.json",
    "color_canary": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-deep-gate-color-repeat256-20260615a13deep2.json",
    "quality_suite": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-deep-gate-quality-suite-20260615a13deep2.json",
    "log": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-deep-gate-20260615a13deep2.log"
  },
  "artifact_exists": {
    "metrics": true,
    "json_canary": true,
    "color_canary": true,
    "quality_suite": true,
    "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_GREEDY_SAMPLE_BF16_TOPK_FASTPATH": null,
    "VLLM_XPU_ASYNC_CLONE_SAMPLED_TOKEN_IDS": null,
    "VLLM_XPU_SANITIZE_LOGITS_NAN": null,
    "VLLM_XPU_LOCAL_ARGMAX_DECODE": null,
    "VLLM_XPU_LOCAL_ARGMAX_DIRECT_GATHER": null,
    "VLLM_XPU_LOCAL_ARGMAX_DIRECT_GATHER_REUSE": null,
    "VLLM_XPU_LOCAL_ARGMAX_PACKED_GATHER": null,
    "VLLM_XPU_LOCAL_ARGMAX_SYNC_TOKENS": 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_CUDAGRAPH_STRONG_OUTPUT": null,
    "VLLM_XPU_CUDAGRAPH_PER_WRAPPER_POOL": null,
    "VLLM_XPU_CUDAGRAPH_NO_GLOBAL_POOL": 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_SANITIZE_REPLAY_INPUTS": 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_STATIC_INPUT_REGEX": null,
    "VLLM_XPU_CUDAGRAPH_STATIC_INPUT_INDICES": null,
    "VLLM_XPU_CUDAGRAPH_STATIC_INPUT_MAX_NUMEL": null,
    "VLLM_XPU_CUDAGRAPH_COMPARE_DIRECT_REGEX": null,
    "VLLM_XPU_CUDAGRAPH_COMPARE_DIRECT_RETURN_DIRECT": null,
    "VLLM_XPU_W8A8_EXPERIMENTAL_ALLOW": null,
    "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",
    "VLLM_XPU_SHARED_EXPERTS_STREAM": null,
    "VLLM_XPU_SHARED_EXPERT_FUSED_ACT_QUANT": null,
    "VLLM_XPU_MOE_SHARED_ADD_ALLREDUCE_CUSTOM_OP": null,
    "GPU_MEMORY_UTILIZATION": "0.90",
    "COMPILATION_CONFIG": "{\"cudagraph_mode\":\"PIECEWISE\"}",
    "ABLATION_FAST_GRAPH_AUTOCONFIG": "1",
    "VLLM_EXTRA_ARGS": "--uvicorn-log-level warning",
    "METRICS_PROMPT_KIND": null,
    "METRICS_PROMPT_PRESET": null,
    "METRICS_RANDOM_PREFIX_LEN": null,
    "METRICS_PROMPT_TOKENS": null,
    "METRICS_OUTPUT_TOKENS": null,
    "METRICS_WARMUP_OUTPUT_TOKENS": null,
    "JSON_LOGPROBS": null,
    "COLOR_LOGPROBS": null
  },
  "metrics_summary": {
    "tok_s_out_client_after_first_chunk": {
      "mean": 93.73361582399541,
      "median": 93.8123812996345,
      "min": 93.34680741514937,
      "max": 93.9628932815633
    },
    "tok_s_out_client_after_first_chunk_corrected": {
      "mean": 93.55054235558917,
      "median": 93.62915399240865,
      "min": 93.16448943191666,
      "max": 93.77937200562275
    },
    "tok_s_out_client_e2e": {
      "mean": 90.62548580766561,
      "median": 90.69861596821744,
      "min": 90.26063583953854,
      "max": 90.84407545468902
    },
    "tok_s_total_client": {
      "mean": 178.77293098777787,
      "median": 178.91719165605394,
      "min": 178.05320741783967,
      "max": 179.2041332211639
    },
    "ttft_ms_client": {
      "mean": 187.33663426246494,
      "median": 187.3550059972331,
      "min": 187.0712200179696,
      "max": 187.56530503742397
    },
    "ttft_ms_vllm_metrics": {
      "mean": 175.54545402526855,
      "median": 175.4699945449829,
      "min": 175.3098964691162,
      "max": 175.9319305419922
    },
    "tok_s_prefill_lower_bound_from_ttft": {
      "mean": 2658.3197710287236,
      "median": 2658.0580483469225,
      "min": 2655.075254459435,
      "max": 2662.0877329616137
    },
    "e2e_ms_vllm_metrics": {
      "mean": 5648.540914058685,
      "median": 5644.018769264221,
      "min": 5634.709596633911,
      "max": 5671.416521072388
    },
    "queue_ms_vllm_histogram": {
      "mean": 0.013698299881070852,
      "median": 0.013736076653003693,
      "min": 0.013335142284631729,
      "max": 0.013985903933644295
    },
    "prefill_ms_vllm_histogram": {
      "mean": 170.84795649861917,
      "median": 170.90937856119126,
      "min": 170.28925591148436,
      "max": 171.2838129606098
    },
    "decode_ms_vllm_histogram": {
      "mean": 5473.221305699553,
      "median": 5468.760248855688,
      "min": 5459.646949078888,
      "max": 5495.71777600795
    },
    "decode_ms_per_generation_token_vllm_histogram": {
      "mean": 10.68988536269444,
      "median": 10.681172361046265,
      "min": 10.663372947419703,
      "max": 10.733823781265528
    },
    "inference_ms_vllm_histogram": {
      "mean": 5644.0692621981725,
      "median": 5639.669627416879,
      "min": 5629.936204990372,
      "max": 5667.00158896856
    },
    "time_per_output_token_ms_vllm_histogram": {
      "mean": 10.710804903521634,
      "median": 10.702074850989604,
      "min": 10.684240604851055,
      "max": 10.75482930725627
    },
    "inter_token_ms_vllm_histogram": {
      "mean": 10.71080490352163,
      "median": 10.702074850989604,
      "min": 10.684240604851052,
      "max": 10.754829307256262
    },
    "iteration_tokens_per_step_vllm_histogram": {
      "mean": 1.97265625,
      "median": 1.97265625,
      "min": 1.97265625,
      "max": 1.97265625
    }
  },
  "speed": {
    "tok_s_out_corrected_mean": 93.55054235558917,
    "tok_s_out_after_first_chunk_mean": 93.73361582399541,
    "tok_s_out_e2e_mean": 90.62548580766561,
    "decode_ms_per_token_mean": 10.68988536269444,
    "ttft_ms_client_mean": 187.33663426246494
  },
  "json_canary": {
    "pass_all": true,
    "repeats_completed": 128,
    "mismatch_count": 0,
    "first_mismatch": null
  },
  "color_canary": {
    "pass_all": true,
    "repeats_completed": 256,
    "mismatch_count": 0,
    "first_mismatch": null
  },
  "quality_suite": {
    "pass_all": true,
    "baseline_match_all": true,
    "exact": {
      "exact_ok": true,
      "copy_phrase": true,
      "arithmetic": true,
      "json_schema": true
    },
    "repeat_pass": true,
    "long_context_pass": true
  },
  "decision": {
    "accepted_by_requested_gates": true,
    "gate_failures": []
  }
}
