{
  "base_url": "http://127.0.0.1:39267/v1",
  "completed": 5,
  "concurrency": 1,
  "context_distribution": {
    "4096": 5
  },
  "context_seed": 0,
  "context_tokens": 4096,
  "engine": "sglang-5f55db35-dflash2-k4",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-05T03:09:23Z",
  "gpu": "RTX PRO 6000 Blackwell Max-Q TP1",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 512,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 89,
    "content_chunks_s": null,
    "decode_tok_s_p50": 86.53957968501697,
    "decode_tok_s_p95": 102.10341196270855,
    "e2e_p50_ms": 697.091699927114,
    "e2e_p95_ms": 1237.561099929735,
    "effective_prefill_tok_s_p50": 34391.70565354196,
    "effective_prefill_tok_s_p95": 34693.88144908757,
    "generation_p50_ms": 586.7200999055058,
    "generation_p95_ms": 618.6150000430644,
    "mean_inter_token_latency_ms_p50": 11.55540625040885,
    "mean_inter_token_latency_ms_p95": 12.257216665602755,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 254,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 5,
    "prompt_tokens": 18065,
    "prompt_tokens_p50": 3613.0,
    "prompt_tokens_p95": 3613.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 64.92203272963954,
    "time_to_first_output_p50_ms": 105.0543999299407,
    "time_to_first_output_p95_ms": 650.8410000242293,
    "ttft_p50_ms": 105.0545999314636,
    "ttft_p95_ms": 650.841600028798,
    "ttft_samples": 5
  },
  "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k4",
  "request_timings": [
    {
      "content_chunks": 18,
      "decode_tok_s": 81.8107305472075,
      "e2e_ms": 1237.561099929735,
      "effective_prefill_tok_s": 5551.279037223372,
      "generation_ms": 586.7200999055058,
      "mean_inter_token_latency_ms": 12.223335414698036,
      "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": 650.8410000242293,
      "ttft_ms": 650.841600028798,
      "visible_generation_ms": 586.7194999009371
    },
    {
      "content_chunks": 18,
      "decode_tok_s": 81.58459030966509,
      "e2e_ms": 697.091699927114,
      "effective_prefill_tok_s": 33224.42441857165,
      "generation_ms": 588.3463999489322,
      "mean_inter_token_latency_ms": 12.257216665602755,
      "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": 108.74529997818172,
      "ttft_ms": 108.74559998046607,
      "visible_generation_ms": 588.3460999466479
    },
    {
      "content_chunks": 18,
      "decode_tok_s": 86.53957968501697,
      "e2e_ms": 658.7989000836387,
      "effective_prefill_tok_s": 34693.88144908757,
      "generation_ms": 554.6595000196248,
      "mean_inter_token_latency_ms": 11.55540625040885,
      "output_token_source": "usage",
      "output_tokens": 49,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 104.1394000640139,
      "ttft_ms": 104.13970006629825,
      "visible_generation_ms": 554.6592000173405
    },
    {
      "content_chunks": 19,
      "decode_tok_s": 88.90828705442193,
      "e2e_ms": 722.9312000563368,
      "effective_prefill_tok_s": 34635.080644619986,
      "generation_ms": 618.6150000430644,
      "mean_inter_token_latency_ms": 11.247545455328442,
      "output_token_source": "usage",
      "output_tokens": 56,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 104.31620001327246,
      "ttft_ms": 104.31640001479536,
      "visible_generation_ms": 618.6148000415415
    },
    {
      "content_chunks": 16,
      "decode_tok_s": 102.10341196270855,
      "e2e_ms": 594.7539999615401,
      "effective_prefill_tok_s": 34391.70565354196,
      "generation_ms": 489.6996000315994,
      "mean_inter_token_latency_ms": 9.793992000631988,
      "output_token_source": "usage",
      "output_tokens": 51,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 105.0543999299407,
      "ttft_ms": 105.0545999314636,
      "visible_generation_ms": 489.6994000300765
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260905T030919Z",
  "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": "2026-09-05T03:09:19Z",
  "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",
    "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",
    "ttft": "request start through the first non-empty streamed content delta"
  },
  "wall_clock_ms": 3912.3851999174803
}