{
  "base_url": "http://127.0.0.1:8081/v1",
  "completed": 5,
  "concurrency": 1,
  "context_distribution": {
    "4096": 5
  },
  "context_seed": 38027,
  "context_tokens": 4096,
  "engine": "ninfer-e3aeaf8c",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-03T12:24:07Z",
  "gpu": "rtx-5090",
  "manifest": null,
  "max_context_tokens": 252928,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 241,
    "content_chunks_s": null,
    "decode_tok_s_p50": 75.25865647689011,
    "decode_tok_s_p95": 75.34204084374771,
    "e2e_p50_ms": 1085.403300006874,
    "e2e_p95_ms": 1226.0135000105947,
    "effective_prefill_tok_s_p50": 8579.636743487514,
    "effective_prefill_tok_s_p95": 8635.749901923871,
    "generation_p50_ms": 623.8216999918222,
    "generation_p95_ms": 704.2545999865979,
    "mean_inter_token_latency_ms_p50": 13.287507999921218,
    "mean_inter_token_latency_ms_p95": 13.309782607551744,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 246,
    "prefix_cache_hit_avg": 0.0,
    "prompt_token_samples": 5,
    "prompt_tokens": 18065,
    "prompt_tokens_p50": 3613.0,
    "prompt_tokens_p95": 3613.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 44.82755196124471,
    "time_to_first_output_p50_ms": 421.11340002156794,
    "time_to_first_output_p95_ms": 602.1918000187725,
    "ttft_p50_ms": 421.1138999671675,
    "ttft_p95_ms": 602.1929999697022,
    "ttft_samples": 5
  },
  "model": "qwen38-27b-ninfer-nvfp4-nospec-252k",
  "request_timings": [
    {
      "content_chunks": 47,
      "decode_tok_s": 75.34204084374771,
      "e2e_ms": 1226.0135000105947,
      "effective_prefill_tok_s": 5999.749581258612,
      "generation_ms": 623.8216999918222,
      "mean_inter_token_latency_ms": 13.27280212748558,
      "output_token_source": "usage",
      "output_tokens": 48,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 602.1918000187725,
      "ttft_ms": 602.1929999697022,
      "visible_generation_ms": 623.8205000408925
    },
    {
      "content_chunks": 45,
      "decode_tok_s": 75.33553190302497,
      "e2e_ms": 1015.7047999673523,
      "effective_prefill_tok_s": 8635.749901923871,
      "generation_ms": 597.3277000011876,
      "mean_inter_token_latency_ms": 13.27394888891528,
      "output_token_source": "usage",
      "output_tokens": 46,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 418.3770999661647,
      "ttft_ms": 418.3779999730177,
      "visible_generation_ms": 597.3267999943346
    },
    {
      "content_chunks": 50,
      "decode_tok_s": 75.25865647689011,
      "e2e_ms": 1085.403300006874,
      "effective_prefill_tok_s": 8581.379048531486,
      "generation_ms": 664.3753999960609,
      "mean_inter_token_latency_ms": 13.287507999921218,
      "output_token_source": "usage",
      "output_tokens": 51,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 421.027900010813,
      "ttft_ms": 421.0287000169046,
      "visible_generation_ms": 664.3745999899693
    },
    {
      "content_chunks": 53,
      "decode_tok_s": 75.25687443292327,
      "e2e_ms": 1125.3680000081658,
      "effective_prefill_tok_s": 8579.636743487514,
      "generation_ms": 704.2545999865979,
      "mean_inter_token_latency_ms": 13.287822641256565,
      "output_token_source": "usage",
      "output_tokens": 54,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 421.11340002156794,
      "ttft_ms": 421.1138999671675,
      "visible_generation_ms": 704.2541000409983
    },
    {
      "content_chunks": 46,
      "decode_tok_s": 75.13270723389704,
      "e2e_ms": 1033.7891999515705,
      "effective_prefill_tok_s": 8570.970386536022,
      "generation_ms": 612.2499999473803,
      "mean_inter_token_latency_ms": 13.309782607551744,
      "output_token_source": "usage",
      "output_tokens": 47,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 421.5392000041902,
      "ttft_ms": 421.53970000799745,
      "visible_generation_ms": 612.249499943573
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260903T122401Z",
  "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": "2026-09-03T12:24:01Z",
  "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": 5487.696499971207
}