{
  "base_url": "http://127.0.0.1:39269/v1",
  "completed": 5,
  "concurrency": 1,
  "context_distribution": {
    "4096": 5
  },
  "context_seed": 0,
  "context_tokens": 4096,
  "engine": "sglang-5f55db35-dflash2-k16",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-05T03:18:16Z",
  "gpu": "RTX PRO 6000 Blackwell Max-Q TP1",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 512,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 71,
    "content_chunks_s": null,
    "decode_tok_s_max": 127.92474758158772,
    "decode_tok_s_mean": 109.21503580324702,
    "decode_tok_s_min": 90.01911218550087,
    "decode_tok_s_p25": 90.27750741476675,
    "decode_tok_s_p50": 112.21301233801422,
    "decode_tok_s_p75": 125.64079949636559,
    "decode_tok_s_p90": 127.92474758158772,
    "decode_tok_s_p95": 127.92474758158772,
    "decode_tok_s_p99": 127.92474758158772,
    "decode_tok_s_samples": 5,
    "e2e_max_ms": 763.27949995175,
    "e2e_mean_ms": 585.4776800144464,
    "e2e_min_ms": 471.43860009964556,
    "e2e_p25_ms": 506.28440000582486,
    "e2e_p50_ms": 543.255600030534,
    "e2e_p75_ms": 643.1302999844775,
    "e2e_p90_ms": 763.27949995175,
    "e2e_p95_ms": 763.27949995175,
    "e2e_p99_ms": 763.27949995175,
    "e2e_samples": 5,
    "effective_prefill_tok_s_max": 34728.66366310464,
    "effective_prefill_tok_s_mean": 30021.270312309352,
    "effective_prefill_tok_s_min": 15704.639760437269,
    "effective_prefill_tok_s_p25": 32422.08112101531,
    "effective_prefill_tok_s_p50": 33353.48881257941,
    "effective_prefill_tok_s_p75": 33897.47820441012,
    "effective_prefill_tok_s_p90": 34728.66366310464,
    "effective_prefill_tok_s_p95": 34728.66366310464,
    "effective_prefill_tok_s_p99": 34728.66366310464,
    "effective_prefill_tok_s_samples": 5,
    "generation_max_ms": 533.2200999837369,
    "generation_mean_ms": 453.3893800107762,
    "generation_min_ms": 367.40350001491606,
    "generation_p25_ms": 397.95989997219294,
    "generation_p50_ms": 436.66950007900596,
    "generation_p75_ms": 531.6939000040293,
    "generation_p90_ms": 533.2200999837369,
    "generation_p95_ms": 533.2200999837369,
    "generation_p99_ms": 533.2200999837369,
    "generation_samples": 5,
    "mean_inter_token_latency_ms_max": 11.108752082994519,
    "mean_inter_token_latency_ms_mean": 9.374724905622498,
    "mean_inter_token_latency_ms_min": 7.817095744998214,
    "mean_inter_token_latency_ms_p25": 7.959197999443859,
    "mean_inter_token_latency_ms_p50": 8.911622450591958,
    "mean_inter_token_latency_ms_p75": 11.076956250083944,
    "mean_inter_token_latency_ms_p90": 11.108752082994519,
    "mean_inter_token_latency_ms_p95": 11.108752082994519,
    "mean_inter_token_latency_ms_p99": 11.108752082994519,
    "mean_inter_token_latency_ms_samples": 5,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 247,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 5,
    "prompt_tokens": 18065,
    "prompt_tokens_max": 3613.0,
    "prompt_tokens_mean": 3613.0,
    "prompt_tokens_min": 3613.0,
    "prompt_tokens_p25": 3613.0,
    "prompt_tokens_p50": 3613.0,
    "prompt_tokens_p75": 3613.0,
    "prompt_tokens_p90": 3613.0,
    "prompt_tokens_p95": 3613.0,
    "prompt_tokens_p99": 3613.0,
    "prompt_tokens_samples": 5,
    "reasoning_chunks": 0,
    "throughput_tok_s": 84.33806027195823,
    "time_to_first_output_max_ms": 230.05939996801317,
    "time_to_first_output_mean_ms": 132.08830000367016,
    "time_to_first_output_min_ms": 104.03510008472949,
    "time_to_first_output_p25_ms": 106.58609995152801,
    "time_to_first_output_p50_ms": 108.32450003363192,
    "time_to_first_output_p75_ms": 111.43639998044819,
    "time_to_first_output_p90_ms": 230.05939996801317,
    "time_to_first_output_p95_ms": 230.05939996801317,
    "time_to_first_output_p99_ms": 230.05939996801317,
    "time_to_first_output_samples": 5,
    "tpot_max_ms": 11.108752082994519,
    "tpot_mean_ms": 9.374724905622498,
    "tpot_min_ms": 7.817095744998214,
    "tpot_p25_ms": 7.959197999443859,
    "tpot_p50_ms": 8.911622450591958,
    "tpot_p75_ms": 11.076956250083944,
    "tpot_p90_ms": 11.108752082994519,
    "tpot_p95_ms": 11.108752082994519,
    "tpot_p99_ms": 11.108752082994519,
    "tpot_samples": 5,
    "ttft_max_ms": 230.0598999718204,
    "ttft_mean_ms": 132.0886200061068,
    "ttft_min_ms": 104.03530008625239,
    "ttft_p25_ms": 106.58639995381236,
    "ttft_p50_ms": 108.32480003591627,
    "ttft_p75_ms": 111.43669998273253,
    "ttft_p90_ms": 230.0598999718204,
    "ttft_p95_ms": 230.0598999718204,
    "ttft_p99_ms": 230.0598999718204,
    "ttft_samples": 5
  },
  "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k16",
  "request_timings": [
    {
      "content_chunks": 16,
      "decode_tok_s": 90.01911218550087,
      "e2e_ms": 763.27949995175,
      "effective_prefill_tok_s": 15704.639760437269,
      "generation_ms": 533.2200999837369,
      "mean_inter_token_latency_ms": 11.108752082994519,
      "output_token_source": "usage",
      "output_tokens": 49,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 230.05939996801317,
      "tpot_ms": 11.108752082994519,
      "ttft_ms": 230.0598999718204,
      "visible_generation_ms": 533.2195999799296
    },
    {
      "content_chunks": 16,
      "decode_tok_s": 90.27750741476675,
      "e2e_ms": 643.1302999844775,
      "effective_prefill_tok_s": 32422.08112101531,
      "generation_ms": 531.6939000040293,
      "mean_inter_token_latency_ms": 11.076956250083944,
      "output_token_source": "usage",
      "output_tokens": 49,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 111.43639998044819,
      "tpot_ms": 11.076956250083944,
      "ttft_ms": 111.43669998273253,
      "visible_generation_ms": 531.6936000017449
    },
    {
      "content_chunks": 14,
      "decode_tok_s": 112.21301233801422,
      "e2e_ms": 543.255600030534,
      "effective_prefill_tok_s": 33897.47820441012,
      "generation_ms": 436.66950007900596,
      "mean_inter_token_latency_ms": 8.911622450591958,
      "output_token_source": "usage",
      "output_tokens": 50,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 106.58609995152801,
      "tpot_ms": 8.911622450591958,
      "ttft_ms": 106.58639995381236,
      "visible_generation_ms": 436.6692000767216
    },
    {
      "content_chunks": 13,
      "decode_tok_s": 125.64079949636559,
      "e2e_ms": 506.28440000582486,
      "effective_prefill_tok_s": 33353.48881257941,
      "generation_ms": 397.95989997219294,
      "mean_inter_token_latency_ms": 7.959197999443859,
      "output_token_source": "usage",
      "output_tokens": 51,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 108.32450003363192,
      "tpot_ms": 7.959197999443859,
      "ttft_ms": 108.32480003591627,
      "visible_generation_ms": 397.9595999699086
    },
    {
      "content_chunks": 12,
      "decode_tok_s": 127.92474758158772,
      "e2e_ms": 471.43860009964556,
      "effective_prefill_tok_s": 34728.66366310464,
      "generation_ms": 367.40350001491606,
      "mean_inter_token_latency_ms": 7.817095744998214,
      "output_token_source": "usage",
      "output_tokens": 48,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 104.03510008472949,
      "tpot_ms": 7.817095744998214,
      "ttft_ms": 104.03530008625239,
      "visible_generation_ms": 367.40330001339316
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260905T031813Z",
  "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": "2026-09-05T03:18:13Z",
  "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": 2928.689599968493
}