{
  "base_url": "http://127.0.0.1:39269/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "32768": 3
  },
  "context_seed": 0,
  "context_tokens": 32768,
  "engine": "sglang-5f55db35-dflash2-k16",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-05T03:18:13Z",
  "gpu": "RTX PRO 6000 Blackwell Max-Q TP1",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 512,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 54,
    "content_chunks_s": null,
    "decode_tok_s_max": 100.79574363637944,
    "decode_tok_s_mean": 88.04147858712065,
    "decode_tok_s_min": 71.13534139701106,
    "decode_tok_s_p25": 71.13534139701106,
    "decode_tok_s_p50": 92.19335072797145,
    "decode_tok_s_p75": 100.79574363637944,
    "decode_tok_s_p90": 100.79574363637944,
    "decode_tok_s_p95": 100.79574363637944,
    "decode_tok_s_p99": 100.79574363637944,
    "decode_tok_s_samples": 3,
    "e2e_max_ms": 4997.665600036271,
    "e2e_mean_ms": 4524.002466661234,
    "e2e_min_ms": 4147.631100029685,
    "e2e_p25_ms": 4147.631100029685,
    "e2e_p50_ms": 4426.710699917749,
    "e2e_p75_ms": 4997.665600036271,
    "e2e_p90_ms": 4997.665600036271,
    "e2e_p95_ms": 4997.665600036271,
    "e2e_p99_ms": 4997.665600036271,
    "e2e_samples": 3,
    "effective_prefill_tok_s_max": 8204.615516666052,
    "effective_prefill_tok_s_mean": 7641.78051915573,
    "effective_prefill_tok_s_min": 6719.544889752884,
    "effective_prefill_tok_s_p25": 6719.544889752884,
    "effective_prefill_tok_s_p50": 8001.181151048252,
    "effective_prefill_tok_s_p75": 8204.615516666052,
    "effective_prefill_tok_s_p90": 8204.615516666052,
    "effective_prefill_tok_s_p95": 8204.615516666052,
    "effective_prefill_tok_s_p99": 8204.615516666052,
    "effective_prefill_tok_s_samples": 3,
    "generation_max_ms": 702.8854999225587,
    "generation_mean_ms": 594.2707666351149,
    "generation_min_ms": 515.8947999589145,
    "generation_p25_ms": 515.8947999589145,
    "generation_p50_ms": 564.0320000238717,
    "generation_p75_ms": 702.8854999225587,
    "generation_p90_ms": 702.8854999225587,
    "generation_p95_ms": 702.8854999225587,
    "generation_p99_ms": 702.8854999225587,
    "generation_samples": 3,
    "mean_inter_token_latency_ms_max": 14.057709998451173,
    "mean_inter_token_latency_ms_mean": 11.608511025014403,
    "mean_inter_token_latency_ms_min": 9.92105384536374,
    "mean_inter_token_latency_ms_p25": 9.92105384536374,
    "mean_inter_token_latency_ms_p50": 10.8467692312283,
    "mean_inter_token_latency_ms_p75": 14.057709998451173,
    "mean_inter_token_latency_ms_p90": 14.057709998451173,
    "mean_inter_token_latency_ms_p95": 14.057709998451173,
    "mean_inter_token_latency_ms_p99": 14.057709998451173,
    "mean_inter_token_latency_ms_samples": 3,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 157,
    "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": 11.565693959360775,
    "time_to_first_output_max_ms": 4433.633600012399,
    "time_to_first_output_mean_ms": 3929.7317000261196,
    "time_to_first_output_min_ms": 3631.73630007077,
    "time_to_first_output_p25_ms": 3631.73630007077,
    "time_to_first_output_p50_ms": 3723.82519999519,
    "time_to_first_output_p75_ms": 4433.633600012399,
    "time_to_first_output_p90_ms": 4433.633600012399,
    "time_to_first_output_p95_ms": 4433.633600012399,
    "time_to_first_output_p99_ms": 4433.633600012399,
    "time_to_first_output_samples": 3,
    "tpot_max_ms": 14.057709998451173,
    "tpot_mean_ms": 11.608511025014403,
    "tpot_min_ms": 9.92105384536374,
    "tpot_p25_ms": 9.92105384536374,
    "tpot_p50_ms": 10.8467692312283,
    "tpot_p75_ms": 14.057709998451173,
    "tpot_p90_ms": 14.057709998451173,
    "tpot_p95_ms": 14.057709998451173,
    "tpot_p99_ms": 14.057709998451173,
    "tpot_samples": 3,
    "ttft_max_ms": 4433.634500019252,
    "ttft_mean_ms": 3929.7323333642757,
    "ttft_min_ms": 3631.736700073816,
    "ttft_p25_ms": 3631.736700073816,
    "ttft_p50_ms": 3723.8257999997586,
    "ttft_p75_ms": 4433.634500019252,
    "ttft_p90_ms": 4433.634500019252,
    "ttft_p95_ms": 4433.634500019252,
    "ttft_p99_ms": 4433.634500019252,
    "ttft_samples": 3
  },
  "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k16",
  "request_timings": [
    {
      "content_chunks": 17,
      "decode_tok_s": 92.19335072797145,
      "e2e_ms": 4997.665600036271,
      "effective_prefill_tok_s": 6719.544889752884,
      "generation_ms": 564.0320000238717,
      "mean_inter_token_latency_ms": 10.8467692312283,
      "output_token_source": "usage",
      "output_tokens": 53,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29792,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 4433.633600012399,
      "tpot_ms": 10.8467692312283,
      "ttft_ms": 4433.634500019252,
      "visible_generation_ms": 564.0311000170186
    },
    {
      "content_chunks": 21,
      "decode_tok_s": 71.13534139701106,
      "e2e_ms": 4426.710699917749,
      "effective_prefill_tok_s": 8001.181151048252,
      "generation_ms": 702.8854999225587,
      "mean_inter_token_latency_ms": 14.057709998451173,
      "output_token_source": "usage",
      "output_tokens": 51,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29795,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 3723.82519999519,
      "tpot_ms": 14.057709998451173,
      "ttft_ms": 3723.8257999997586,
      "visible_generation_ms": 702.88489991799
    },
    {
      "content_chunks": 16,
      "decode_tok_s": 100.79574363637944,
      "e2e_ms": 4147.631100029685,
      "effective_prefill_tok_s": 8204.615516666052,
      "generation_ms": 515.8947999589145,
      "mean_inter_token_latency_ms": 9.92105384536374,
      "output_token_source": "usage",
      "output_tokens": 53,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29797,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 3631.73630007077,
      "tpot_ms": 9.92105384536374,
      "ttft_ms": 3631.736700073816,
      "visible_generation_ms": 515.8943999558687
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260905T031759Z",
  "schema": "anvil-serving.benchmark/v1",
  "serve_flags": {
    "no_thinking": true,
    "reasoning_effort": null,
    "shared_prefix_burst": false,
    "thinking_mode": "disabled"
  },
  "source_recipe": null,
  "started_at": "2026-09-05T03:17:59Z",
  "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": 13574.628600035794
}
