{
  "base_url": "http://127.0.0.1:39275/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "32768": 3
  },
  "context_seed": 0,
  "context_tokens": 32768,
  "engine": "sglang",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-05T03:44:04Z",
  "gpu": "Primary Node / RTX PRO 6000 Blackwell Max-Q / TP=1",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 512,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 60,
    "content_chunks_s": null,
    "decode_tok_s_max": 92.20319357710308,
    "decode_tok_s_mean": 86.45585221029336,
    "decode_tok_s_min": 81.70463093923455,
    "decode_tok_s_p25": 81.70463093923455,
    "decode_tok_s_p50": 85.45973211454246,
    "decode_tok_s_p75": 92.20319357710308,
    "decode_tok_s_p90": 92.20319357710308,
    "decode_tok_s_p95": 92.20319357710308,
    "decode_tok_s_p99": 92.20319357710308,
    "decode_tok_s_samples": 3,
    "e2e_max_ms": 5348.449099925347,
    "e2e_mean_ms": 4871.49733332141,
    "e2e_min_ms": 4344.8155000805855,
    "e2e_p25_ms": 4344.8155000805855,
    "e2e_p50_ms": 4921.227399958298,
    "e2e_p75_ms": 5348.449099925347,
    "e2e_p90_ms": 5348.449099925347,
    "e2e_p95_ms": 5348.449099925347,
    "e2e_p99_ms": 5348.449099925347,
    "e2e_samples": 3,
    "effective_prefill_tok_s_max": 7631.517205515982,
    "effective_prefill_tok_s_mean": 7161.2116666322645,
    "effective_prefill_tok_s_min": 6240.950679168722,
    "effective_prefill_tok_s_p25": 6240.950679168722,
    "effective_prefill_tok_s_p50": 7611.16711521209,
    "effective_prefill_tok_s_p75": 7631.517205515982,
    "effective_prefill_tok_s_p90": 7631.517205515982,
    "effective_prefill_tok_s_p95": 7631.517205515982,
    "effective_prefill_tok_s_p99": 7631.517205515982,
    "effective_prefill_tok_s_samples": 3,
    "generation_max_ms": 1006.3219000585377,
    "generation_mean_ms": 673.9169333595783,
    "generation_min_ms": 440.6115000601858,
    "generation_p25_ms": 440.6115000601858,
    "generation_p50_ms": 574.8173999600112,
    "generation_p75_ms": 1006.3219000585377,
    "generation_p90_ms": 1006.3219000585377,
    "generation_p95_ms": 1006.3219000585377,
    "generation_p99_ms": 1006.3219000585377,
    "generation_samples": 3,
    "mean_inter_token_latency_ms_max": 12.239208335005161,
    "mean_inter_token_latency_ms_mean": 11.595412365848837,
    "mean_inter_token_latency_ms_min": 10.845611320000213,
    "mean_inter_token_latency_ms_p25": 10.845611320000213,
    "mean_inter_token_latency_ms_p50": 11.701417442541135,
    "mean_inter_token_latency_ms_p75": 12.239208335005161,
    "mean_inter_token_latency_ms_p90": 12.239208335005161,
    "mean_inter_token_latency_ms_p95": 12.239208335005161,
    "mean_inter_token_latency_ms_p99": 12.239208335005161,
    "mean_inter_token_latency_ms_samples": 3,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 178,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 3,
    "prompt_tokens": 89384,
    "prompt_tokens_max": 29797.0,
    "prompt_tokens_mean": 29794.666666666668,
    "prompt_tokens_min": 29792.0,
    "prompt_tokens_p25": 29792.0,
    "prompt_tokens_p50": 29795.0,
    "prompt_tokens_p75": 29797.0,
    "prompt_tokens_p90": 29797.0,
    "prompt_tokens_p95": 29797.0,
    "prompt_tokens_p99": 29797.0,
    "prompt_tokens_samples": 3,
    "reasoning_chunks": 0,
    "throughput_tok_s": 12.17800266304636,
    "time_to_first_output_max_ms": 4773.631699965335,
    "time_to_first_output_mean_ms": 4197.580399961832,
    "time_to_first_output_min_ms": 3904.2040000203997,
    "time_to_first_output_p25_ms": 3904.2040000203997,
    "time_to_first_output_p50_ms": 3914.90549989976,
    "time_to_first_output_p75_ms": 4773.631699965335,
    "time_to_first_output_p90_ms": 4773.631699965335,
    "time_to_first_output_p95_ms": 4773.631699965335,
    "time_to_first_output_p99_ms": 4773.631699965335,
    "time_to_first_output_samples": 3,
    "tpot_max_ms": 12.239208335005161,
    "tpot_mean_ms": 11.595412365848837,
    "tpot_min_ms": 10.845611320000213,
    "tpot_p25_ms": 10.845611320000213,
    "tpot_p50_ms": 11.701417442541135,
    "tpot_p75_ms": 12.239208335005161,
    "tpot_p90_ms": 12.239208335005161,
    "tpot_p95_ms": 12.239208335005161,
    "tpot_p99_ms": 12.239208335005161,
    "tpot_samples": 3,
    "ttft_max_ms": 4773.632399970666,
    "ttft_mean_ms": 4197.581166634336,
    "ttft_min_ms": 3904.2051000287756,
    "ttft_p25_ms": 3904.2051000287756,
    "ttft_p50_ms": 3914.905999903567,
    "ttft_p75_ms": 4773.632399970666,
    "ttft_p90_ms": 4773.632399970666,
    "ttft_p95_ms": 4773.632399970666,
    "ttft_p99_ms": 4773.632399970666,
    "ttft_samples": 3
  },
  "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12-chunk1k-lazy96",
  "request_timings": [
    {
      "content_chunks": 17,
      "decode_tok_s": 92.20319357710308,
      "e2e_ms": 5348.449099925347,
      "effective_prefill_tok_s": 6240.950679168722,
      "generation_ms": 574.8173999600112,
      "mean_inter_token_latency_ms": 10.845611320000213,
      "output_token_source": "usage",
      "output_tokens": 54,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29792,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 4773.631699965335,
      "tpot_ms": 10.845611320000213,
      "ttft_ms": 4773.632399970666,
      "visible_generation_ms": 574.8166999546811
    },
    {
      "content_chunks": 13,
      "decode_tok_s": 81.70463093923455,
      "e2e_ms": 4344.8155000805855,
      "effective_prefill_tok_s": 7631.517205515982,
      "generation_ms": 440.6115000601858,
      "mean_inter_token_latency_ms": 12.239208335005161,
      "output_token_source": "usage",
      "output_tokens": 37,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29795,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 3904.2040000203997,
      "tpot_ms": 12.239208335005161,
      "ttft_ms": 3904.2051000287756,
      "visible_generation_ms": 440.61040005180985
    },
    {
      "content_chunks": 30,
      "decode_tok_s": 85.45973211454246,
      "e2e_ms": 4921.227399958298,
      "effective_prefill_tok_s": 7611.16711521209,
      "generation_ms": 1006.3219000585377,
      "mean_inter_token_latency_ms": 11.701417442541135,
      "output_token_source": "usage",
      "output_tokens": 87,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29797,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 3914.90549989976,
      "tpot_ms": 11.701417442541135,
      "ttft_ms": 3914.905999903567,
      "visible_generation_ms": 1006.3214000547305
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260905T034349Z",
  "schema": "anvil-serving.benchmark/v1",
  "serve_flags": {
    "no_thinking": false,
    "reasoning_effort": null,
    "shared_prefix_burst": false,
    "thinking_mode": "disabled"
  },
  "source_recipe": null,
  "started_at": "2026-09-05T03:43:49Z",
  "tier": null,
  "timing_methodology": {
    "clock": "client time.perf_counter",
    "decode": "usage completion tokens after the first token divided by client-observed generation time; completion tokens may include reasoning tokens",
    "distributions": "request-level mean, min, p25, p50, p75, p90, p95, p99, and max using the nearest-rank percentile method",
    "effective_prefill": "usage.prompt_tokens divided by client-observed time to first output; includes queueing, scheduling, prefill, and first-token work",
    "generation": "client-observed E2E minus time to first output",
    "mean_inter_token_latency": "client-observed generation time divided by completion tokens after the first token; not a raw per-token timestamp trace",
    "time_to_first_output": "request start through the first non-empty streamed reasoning or content delta",
    "tpot": "alias of mean_inter_token_latency for capacity-v3: client-observed generation time divided by completion tokens after the first token",
    "ttft": "request start through the first non-empty streamed content delta"
  },
  "wall_clock_ms": 14616.518400027417
}
