{
  "base_url": "http://127.0.0.1:39276/v1",
  "completed": 8,
  "concurrency": 8,
  "context_distribution": {
    "81920": 8
  },
  "context_seed": 0,
  "context_tokens": 81920,
  "engine": "sglang",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-05T03:48:40Z",
  "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": 155,
    "content_chunks_s": null,
    "decode_tok_s_max": 82.64453223844566,
    "decode_tok_s_mean": 11.605057552012175,
    "decode_tok_s_min": 0.303919088892645,
    "decode_tok_s_p25": 0.6485866412908512,
    "decode_tok_s_p50": 0.8374356763802392,
    "decode_tok_s_p75": 2.605102225741617,
    "decode_tok_s_p90": 82.64453223844566,
    "decode_tok_s_p95": 82.64453223844566,
    "decode_tok_s_p99": 82.64453223844566,
    "decode_tok_s_samples": 8,
    "e2e_max_ms": 139973.4088999685,
    "e2e_mean_ms": 137279.92092499335,
    "e2e_min_ms": 120952.94119999744,
    "e2e_p25_ms": 139315.21480006631,
    "e2e_p50_ms": 139446.85649999883,
    "e2e_p75_ms": 139798.47659997176,
    "e2e_p90_ms": 139973.4088999685,
    "e2e_p95_ms": 139973.4088999685,
    "e2e_p99_ms": 139973.4088999685,
    "e2e_samples": 8,
    "effective_prefill_tok_s_max": 4416.321420306757,
    "effective_prefill_tok_s_mean": 1521.0465178548682,
    "effective_prefill_tok_s_min": 557.5407683739377,
    "effective_prefill_tok_s_p25": 646.8851216424996,
    "effective_prefill_tok_s_p50": 905.5973019664414,
    "effective_prefill_tok_s_p75": 1505.9080070471075,
    "effective_prefill_tok_s_p90": 4416.321420306757,
    "effective_prefill_tok_s_p95": 4416.321420306757,
    "effective_prefill_tok_s_p99": 4416.321420306757,
    "effective_prefill_tok_s_samples": 8,
    "generation_max_ms": 121742.92879994027,
    "generation_mean_ms": 59762.684012486716,
    "generation_min_ms": 689.7007999941707,
    "generation_p25_ms": 18041.51849995833,
    "generation_p50_ms": 53735.47039995901,
    "generation_p75_ms": 88258.14369996078,
    "generation_p90_ms": 121742.92879994027,
    "generation_p95_ms": 121742.92879994027,
    "generation_p99_ms": 121742.92879994027,
    "generation_samples": 8,
    "mean_inter_token_latency_ms_max": 3290.3494270254128,
    "mean_inter_token_latency_ms_mean": 1164.2433086024855,
    "mean_inter_token_latency_ms_min": 12.10001403498545,
    "mean_inter_token_latency_ms_p25": 234.94881058851365,
    "mean_inter_token_latency_ms_p50": 1117.1916924045668,
    "mean_inter_token_latency_ms_p75": 1539.558808696355,
    "mean_inter_token_latency_ms_p90": 3290.3494270254128,
    "mean_inter_token_latency_ms_p95": 3290.3494270254128,
    "mean_inter_token_latency_ms_p99": 3290.3494270254128,
    "mean_inter_token_latency_ms_samples": 8,
    "output_token_sources": {
      "usage": 8
    },
    "output_tokens": 472,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 8,
    "prompt_tokens": 620979,
    "prompt_tokens_max": 77631.0,
    "prompt_tokens_mean": 77622.375,
    "prompt_tokens_min": 77615.0,
    "prompt_tokens_p25": 77617.0,
    "prompt_tokens_p50": 77620.0,
    "prompt_tokens_p75": 77626.0,
    "prompt_tokens_p90": 77631.0,
    "prompt_tokens_p95": 77631.0,
    "prompt_tokens_p99": 77631.0,
    "prompt_tokens_samples": 8,
    "reasoning_chunks": 0,
    "throughput_tok_s": 3.371870551404339,
    "time_to_first_output_max_ms": 139238.2484000409,
    "time_to_first_output_mean_ms": 77517.23691250663,
    "time_to_first_output_min_ms": 17575.03419998102,
    "time_to_first_output_p25_ms": 34471.85900004115,
    "time_to_first_output_p50_ms": 68686.85199995525,
    "time_to_first_output_p75_ms": 102911.42270003911,
    "time_to_first_output_p90_ms": 139238.2484000409,
    "time_to_first_output_p95_ms": 139238.2484000409,
    "time_to_first_output_p99_ms": 139238.2484000409,
    "time_to_first_output_samples": 8,
    "tpot_max_ms": 3290.3494270254128,
    "tpot_mean_ms": 1164.2433086024855,
    "tpot_min_ms": 12.10001403498545,
    "tpot_p25_ms": 234.94881058851365,
    "tpot_p50_ms": 1117.1916924045668,
    "tpot_p75_ms": 1539.558808696355,
    "tpot_p90_ms": 3290.3494270254128,
    "tpot_p95_ms": 3290.3494270254128,
    "tpot_p99_ms": 3290.3494270254128,
    "tpot_samples": 8,
    "ttft_max_ms": 139238.24880004395,
    "ttft_mean_ms": 77517.23736251006,
    "ttft_min_ms": 17575.03499998711,
    "ttft_p25_ms": 34471.85960004572,
    "ttft_p50_ms": 68686.8523999583,
    "ttft_p75_ms": 102911.4230000414,
    "ttft_p90_ms": 139238.24880004395,
    "ttft_p95_ms": 139238.24880004395,
    "ttft_p99_ms": 139238.24880004395,
    "ttft_samples": 8
  },
  "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12-chunk1k-mamba40",
  "request_timings": [
    {
      "content_chunks": 26,
      "decode_tok_s": 0.8951015361094107,
      "e2e_ms": 139798.47659997176,
      "effective_prefill_tok_s": 1505.9080070471075,
      "generation_ms": 88258.14369996078,
      "mean_inter_token_latency_ms": 1117.1916924045668,
      "output_token_source": "usage",
      "output_tokens": 80,
      "planned_context_tokens": 81920,
      "prompt_tokens": 77615,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 51540.33290001098,
      "tpot_ms": 1117.1916924045668,
      "ttft_ms": 51540.333200013265,
      "visible_generation_ms": 88258.1433999585
    },
    {
      "content_chunks": 16,
      "decode_tok_s": 0.303919088892645,
      "e2e_ms": 139317.9629999213,
      "effective_prefill_tok_s": 4416.321420306757,
      "generation_ms": 121742.92879994027,
      "mean_inter_token_latency_ms": 3290.3494270254128,
      "output_token_source": "usage",
      "output_tokens": 38,
      "planned_context_tokens": 81920,
      "prompt_tokens": 77617,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 17575.03419998102,
      "tpot_ms": 3290.3494270254128,
      "ttft_ms": 17575.03499998711,
      "visible_generation_ms": 121742.92799993418
    },
    {
      "content_chunks": 20,
      "decode_tok_s": 0.6495367337391712,
      "e2e_ms": 139506.55719998758,
      "effective_prefill_tok_s": 1130.0415980637833,
      "generation_ms": 70819.70520003233,
      "mean_inter_token_latency_ms": 1539.558808696355,
      "output_token_source": "usage",
      "output_tokens": 47,
      "planned_context_tokens": 81920,
      "prompt_tokens": 77619,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 68686.85199995525,
      "tpot_ms": 1539.558808696355,
      "ttft_ms": 68686.8523999583,
      "visible_generation_ms": 70819.70480002929
    },
    {
      "content_chunks": 18,
      "decode_tok_s": 0.8374356763802392,
      "e2e_ms": 139446.85649999883,
      "effective_prefill_tok_s": 905.5973019664414,
      "generation_ms": 53735.47039995901,
      "mean_inter_token_latency_ms": 1194.1215644435335,
      "output_token_source": "usage",
      "output_tokens": 46,
      "planned_context_tokens": 81920,
      "prompt_tokens": 77620,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 85711.38610003982,
      "tpot_ms": 1194.1215644435335,
      "ttft_ms": 85711.38650004286,
      "visible_generation_ms": 53735.46999995597
    },
    {
      "content_chunks": 17,
      "decode_tok_s": 0.6485866412908512,
      "e2e_ms": 139315.21480006631,
      "effective_prefill_tok_s": 2251.778762494571,
      "generation_ms": 104843.35580002517,
      "mean_inter_token_latency_ms": 1541.814055882723,
      "output_token_source": "usage",
      "output_tokens": 69,
      "planned_context_tokens": 81920,
      "prompt_tokens": 77623,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 34471.85900004115,
      "tpot_ms": 1541.814055882723,
      "ttft_ms": 34471.85960004572,
      "visible_generation_ms": 104843.3552000206
    },
    {
      "content_chunks": 14,
      "decode_tok_s": 2.605102225741617,
      "e2e_ms": 120952.94119999744,
      "effective_prefill_tok_s": 754.2991629438478,
      "generation_ms": 18041.51849995833,
      "mean_inter_token_latency_ms": 383.8620957437942,
      "output_token_source": "usage",
      "output_tokens": 48,
      "planned_context_tokens": 81920,
      "prompt_tokens": 77626,
      "reasoning_chunks": 0,
      "request_index": 5,
      "time_to_first_output_ms": 102911.42270003911,
      "tpot_ms": 383.8620957437942,
      "ttft_ms": 102911.4230000414,
      "visible_generation_ms": 18041.518199956045
    },
    {
      "content_chunks": 30,
      "decode_tok_s": 4.256246275497802,
      "e2e_ms": 139973.4088999685,
      "effective_prefill_tok_s": 646.8851216424996,
      "generation_ms": 19970.64890002366,
      "mean_inter_token_latency_ms": 234.94881058851365,
      "output_token_source": "usage",
      "output_tokens": 86,
      "planned_context_tokens": 81920,
      "prompt_tokens": 77628,
      "reasoning_chunks": 0,
      "request_index": 6,
      "time_to_first_output_ms": 120002.75999994483,
      "tpot_ms": 234.94881058851365,
      "ttft_ms": 120002.76039994787,
      "visible_generation_ms": 19970.648500020616
    },
    {
      "content_chunks": 14,
      "decode_tok_s": 82.64453223844566,
      "e2e_ms": 139927.94920003507,
      "effective_prefill_tok_s": 557.5407683739377,
      "generation_ms": 689.7007999941707,
      "mean_inter_token_latency_ms": 12.10001403498545,
      "output_token_source": "usage",
      "output_tokens": 58,
      "planned_context_tokens": 81920,
      "prompt_tokens": 77631,
      "reasoning_chunks": 0,
      "request_index": 7,
      "time_to_first_output_ms": 139238.2484000409,
      "tpot_ms": 12.10001403498545,
      "ttft_ms": 139238.24880004395,
      "visible_generation_ms": 689.7003999911249
    }
  ],
  "requests": 8,
  "run_id": "benchmark-20260905T034620Z",
  "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": "2026-09-05T03:46:20Z",
  "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": 139981.64899996482
}
