{
  "label": "prefill-safe-int8-mixed-workspace-async-tp2-smoke",
  "stamp": "20260615tp2safe1",
  "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-tp2-smoke-p512o512-20260615tp2safe1.json",
    "json_canary": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-tp2-smoke-json-repeat16-20260615tp2safe1.json",
    "color_canary": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-tp2-smoke-color-repeat16-20260615tp2safe1.json",
    "quality_suite": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-tp2-smoke-quality-suite-20260615tp2safe1.json",
    "log": "/home/steve/llm-optimizations/data/qwen36-ablation-prefill-safe-int8-mixed-workspace-async-tp2-smoke-20260615tp2safe1.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_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_INT8_MOE_PERSISTENT_SCRATCH": null,
    "VLLM_XPU_INT8_MOE_FUSED_PROLOGUE_OFFSET": null,
    "VLLM_XPU_INT8_MOE_FUSED_PROLOGUE_OFFSET_ALLOW_CAPTURE": null,
    "VLLM_XPU_SHARED_EXPERTS_STREAM": null,
    "VLLM_XPU_SHARED_EXPERT_FUSED_ACT_QUANT": null,
    "VLLM_XPU_MOE_SHARED_ADD_ALLREDUCE_CUSTOM_OP": null,
    "VLLM_XPU_METADATA_COPY_ALLOW": null,
    "VLLM_XPU_BLOCK_TABLE_DIRTY_COMMIT": null,
    "VLLM_XPU_BLOCK_TABLE_DIRTY_COMMIT_LOG_EVERY": 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": 86.03715537909211,
      "median": 86.03715537909211,
      "min": 86.03715537909211,
      "max": 86.03715537909211
    },
    "tok_s_out_client_after_first_chunk_corrected": {
      "mean": 85.86911405999231,
      "median": 85.86911405999231,
      "min": 85.86911405999231,
      "max": 85.86911405999231
    },
    "tok_s_out_client_e2e": {
      "mean": 82.26607414527504,
      "median": 82.26607414527504,
      "min": 82.26607414527504,
      "max": 82.26607414527504
    },
    "tok_s_total_client": {
      "mean": 162.28268532564022,
      "median": 162.28268532564022,
      "min": 162.28268532564022,
      "max": 162.28268532564022
    },
    "ttft_ms_client": {
      "mean": 272.79036887921393,
      "median": 272.79036887921393,
      "min": 272.79036887921393,
      "max": 272.79036887921393
    },
    "ttft_ms_vllm_metrics": {
      "mean": 260.2367401123047,
      "median": 260.2367401123047,
      "min": 260.2367401123047,
      "max": 260.2367401123047
    },
    "tok_s_prefill_lower_bound_from_ttft": {
      "mean": 1825.5776479429314,
      "median": 1825.5776479429314,
      "min": 1825.5776479429314,
      "max": 1825.5776479429314
    },
    "e2e_ms_vllm_metrics": {
      "mean": 6222.656726837158,
      "median": 6222.656726837158,
      "min": 6222.656726837158,
      "max": 6222.656726837158
    },
    "queue_ms_vllm_histogram": {
      "mean": 0.014377059414982796,
      "median": 0.014377059414982796,
      "min": 0.014377059414982796,
      "max": 0.014377059414982796
    },
    "prefill_ms_vllm_histogram": {
      "mean": 255.0938909407705,
      "median": 255.0938909407705,
      "min": 255.0938909407705,
      "max": 255.0938909407705
    },
    "decode_ms_vllm_histogram": {
      "mean": 5962.670223088935,
      "median": 5962.670223088935,
      "min": 5962.670223088935,
      "max": 5962.670223088935
    },
    "decode_ms_per_generation_token_vllm_histogram": {
      "mean": 11.645840279470576,
      "median": 11.645840279470576,
      "min": 11.645840279470576,
      "max": 11.645840279470576
    },
    "inference_ms_vllm_histogram": {
      "mean": 6217.7641140297055,
      "median": 6217.7641140297055,
      "min": 6217.7641140297055,
      "max": 6217.7641140297055
    },
    "time_per_output_token_ms_vllm_histogram": {
      "mean": 11.668630573559573,
      "median": 11.668630573559573,
      "min": 11.668630573559573,
      "max": 11.668630573559573
    },
    "inter_token_ms_vllm_histogram": {
      "mean": 11.668630573559561,
      "median": 11.668630573559561,
      "min": 11.668630573559561,
      "max": 11.668630573559561
    },
    "iteration_tokens_per_step_vllm_histogram": {
      "mean": 1.97265625,
      "median": 1.97265625,
      "min": 1.97265625,
      "max": 1.97265625
    }
  },
  "speed": {
    "tok_s_out_corrected_mean": 85.86911405999231,
    "tok_s_out_after_first_chunk_mean": 86.03715537909211,
    "tok_s_out_e2e_mean": 82.26607414527504,
    "decode_ms_per_token_mean": 11.645840279470576,
    "ttft_ms_client_mean": 272.79036887921393
  },
  "json_canary": {
    "pass_all": true,
    "repeats_completed": 16,
    "mismatch_count": 0,
    "first_mismatch": null
  },
  "color_canary": {
    "pass_all": true,
    "repeats_completed": 16,
    "mismatch_count": 0,
    "first_mismatch": null
  },
  "decision": {
    "accepted_by_requested_gates": true,
    "gate_failures": []
  }
}
