{
  "label": "eagle2-tokenheavy-synthaccept6-piecewise-tp2-k5-ceiling-20260618h",
  "stamp": "20260618h02",
  "return_codes": {
    "metrics": 0,
    "json_canary": 0,
    "color_canary": 0,
    "quality_suite": 0
  },
  "status": {
    "metrics": "passed",
    "json_canary": "skipped",
    "color_canary": "skipped",
    "quality_suite": "skipped"
  },
  "artifacts": {
    "metrics": "/home/steve/llm-optimizations/data/qwen36-ablation-eagle2-tokenheavy-synthaccept6-piecewise-tp2-k5-ceiling-20260618h-p512o512-20260618h02.json",
    "json_canary": "/home/steve/llm-optimizations/data/qwen36-ablation-eagle2-tokenheavy-synthaccept6-piecewise-tp2-k5-ceiling-20260618h-json-repeat96-20260618h02.json",
    "color_canary": "/home/steve/llm-optimizations/data/qwen36-ablation-eagle2-tokenheavy-synthaccept6-piecewise-tp2-k5-ceiling-20260618h-color-repeat96-20260618h02.json",
    "quality_suite": "/home/steve/llm-optimizations/data/qwen36-ablation-eagle2-tokenheavy-synthaccept6-piecewise-tp2-k5-ceiling-20260618h-quality-suite-20260618h02.json",
    "log": "/home/steve/llm-optimizations/data/qwen36-ablation-eagle2-tokenheavy-synthaccept6-piecewise-tp2-k5-ceiling-20260618h-20260618h02.log"
  },
  "artifact_exists": {
    "metrics": true,
    "json_canary": false,
    "color_canary": false,
    "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_DUMMY_LOGIT_INDICES_STATIC": 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_INT8_MOE_FUSED_PROLOGUE_OFFSET_SKIP_LAYERS": null,
    "VLLM_XPU_INT8_MOE_FUSED_PROLOGUE_OFFSET_LAYER_REGEX": 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": "0.82",
    "COMPILATION_CONFIG": "{\"cudagraph_mode\":\"PIECEWISE\",\"max_cudagraph_capture_size\":128}",
    "SPEC_CONFIG": "{\"method\":\"eagle\",\"model\":\"/home/steve/llm-optimizations/data/qwen36-eagle2-corpus2-tokenheavy-trained-20260618a\",\"num_speculative_tokens\":5,\"rejection_sample_method\":\"synthetic\",\"synthetic_acceptance_length\":6.0}",
    "ABLATION_FAST_GRAPH_AUTOCONFIG": "1",
    "VLLM_EXTRA_ARGS": null,
    "METRICS_PROMPT_KIND": null,
    "METRICS_PROMPT_PRESET": "natural-chat",
    "METRICS_RANDOM_PREFIX_LEN": null,
    "METRICS_PROMPT_TOKENS": "512",
    "METRICS_OUTPUT_TOKENS": "128",
    "METRICS_WARMUP_OUTPUT_TOKENS": "32",
    "JSON_LOGPROBS": null,
    "COLOR_LOGPROBS": null
  },
  "metrics_summary": {
    "tok_s_out_client_after_first_chunk": {
      "mean": 188.53092245057707,
      "median": 188.53092245057707,
      "min": 186.69180725898775,
      "max": 190.37003764216635
    },
    "tok_s_out_client_after_first_chunk_corrected": {
      "mean": 181.9100662518911,
      "median": 181.9100662518911,
      "min": 179.39915853793354,
      "max": 184.42097396584865
    },
    "tok_s_out_client_e2e": {
      "mean": 86.33955791101282,
      "median": 86.33955791101282,
      "min": 85.97100410865447,
      "max": 86.70811171337117
    },
    "tok_s_total_client": {
      "mean": 422.25440040854704,
      "median": 422.25440040854704,
      "min": 420.45194196888826,
      "max": 424.0568588482058
    },
    "ttft_ms_client": {
      "mean": 803.5472775809467,
      "median": 803.5472775809467,
      "min": 803.2521391287446,
      "max": 803.8424160331488
    },
    "ttft_ms_vllm_metrics": {
      "mean": 770.2072858810425,
      "median": 770.2072858810425,
      "min": 769.9780464172363,
      "max": 770.4365253448486
    },
    "tok_s_prefill_lower_bound_from_ttft": {
      "mean": 619.7520433172482,
      "median": 619.7520433172482,
      "min": 619.5244118338035,
      "max": 619.9796748006929
    },
    "e2e_ms_vllm_metrics": {
      "mean": 1481.4860820770264,
      "median": 1481.4860820770264,
      "min": 1475.149393081665,
      "max": 1487.8227710723877
    },
    "queue_ms_vllm_histogram": {
      "mean": 0.012989388778805733,
      "median": 0.012989388778805733,
      "min": 0.011751893907785416,
      "max": 0.01422688364982605
    },
    "prefill_ms_vllm_histogram": {
      "mean": 764.9570555659011,
      "median": 764.9570555659011,
      "min": 764.9481231346726,
      "max": 764.9659879971296
    },
    "decode_ms_vllm_histogram": {
      "mean": 711.5481980144978,
      "median": 711.5481980144978,
      "min": 704.8777160234749,
      "max": 718.2186800055206
    },
    "decode_ms_per_generation_token_vllm_histogram": {
      "mean": 5.558970296988264,
      "median": 5.558970296988264,
      "min": 5.506857156433398,
      "max": 5.6110834375431295
    },
    "inference_ms_vllm_histogram": {
      "mean": 1476.5052535803989,
      "median": 1476.5052535803989,
      "min": 1469.8258391581476,
      "max": 1483.1846680026501
    },
    "time_per_output_token_ms_vllm_histogram": {
      "mean": 5.602741716649595,
      "median": 5.602741716649595,
      "min": 5.550218236405335,
      "max": 5.655265196893855
    },
    "inter_token_ms_vllm_histogram": {
      "mean": 32.3430999097499,
      "median": 32.3430999097499,
      "min": 32.03989618288523,
      "max": 32.646303636614576
    },
    "iteration_tokens_per_step_vllm_histogram": {
      "mean": 27.217391304347824,
      "median": 27.217391304347824,
      "min": 27.217391304347824,
      "max": 27.217391304347824
    }
  },
  "speed": {
    "tok_s_out_corrected_mean": 181.9100662518911,
    "tok_s_out_after_first_chunk_mean": 188.53092245057707,
    "tok_s_out_e2e_mean": 86.33955791101282,
    "decode_ms_per_token_mean": 5.558970296988264,
    "ttft_ms_client_mean": 803.5472775809467
  },
  "decision": {
    "accepted_by_requested_gates": true,
    "gate_failures": []
  }
}
