{
  "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": "2026-09-03T12:29:04Z",
  "gpu": "rtx-5090",
  "manifest": null,
  "max_context_tokens": 252928,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 91,
    "content_chunks_s": null,
    "decode_tok_s_p50": 184.22869709500787,
    "decode_tok_s_p95": 185.00593500866512,
    "e2e_p50_ms": 730.2486000116915,
    "e2e_p95_ms": 868.070499971509,
    "effective_prefill_tok_s_p50": 8137.409380228871,
    "effective_prefill_tok_s_p95": 8142.973760339138,
    "generation_p50_ms": 271.2319999700412,
    "generation_p95_ms": 287.34759998042136,
    "mean_inter_token_latency_ms_p50": 5.428035999648273,
    "mean_inter_token_latency_ms_p95": 6.027377777112027,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 248,
    "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": 65.93633139914084,
    "time_to_first_output_p50_ms": 443.99880000855774,
    "time_to_first_output_p95_ms": 597.8088999981992,
    "ttft_p50_ms": 443.9991999533959,
    "ttft_p95_ms": 597.8103000088595,
    "ttft_samples": 5
  },
  "model": "qwen38-27b-ninfer-nvfp4-mtp3-252k",
  "request_timings": [
    {
      "content_chunks": 18,
      "decode_tok_s": 185.00593500866512,
      "e2e_ms": 868.070499971509,
      "effective_prefill_tok_s": 6043.737388337449,
      "generation_ms": 270.2615999733098,
      "mean_inter_token_latency_ms": 5.405231999466196,
      "output_token_source": "usage",
      "output_tokens": 51,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 597.8088999981992,
      "ttft_ms": 597.8103000088595,
      "visible_generation_ms": 270.2601999626495
    },
    {
      "content_chunks": 18,
      "decode_tok_s": 165.98085834051668,
      "e2e_ms": 730.2486000116915,
      "effective_prefill_tok_s": 7869.179518846009,
      "generation_ms": 271.11560001503676,
      "mean_inter_token_latency_ms": 6.0247911114452615,
      "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": 459.1329999966547,
      "ttft_ms": 459.1336000012234,
      "visible_generation_ms": 271.11500001046807
    },
    {
      "content_chunks": 19,
      "decode_tok_s": 184.44559830536673,
      "e2e_ms": 731.1450999695808,
      "effective_prefill_tok_s": 8141.1003894529695,
      "generation_ms": 287.34759998042136,
      "mean_inter_token_latency_ms": 5.421652829819271,
      "output_token_source": "usage",
      "output_tokens": 54,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 443.7974999891594,
      "ttft_ms": 443.7978999922052,
      "visible_generation_ms": 287.34719997737557
    },
    {
      "content_chunks": 18,
      "decode_tok_s": 184.22869709500787,
      "e2e_ms": 715.4005999909714,
      "effective_prefill_tok_s": 8137.409380228871,
      "generation_ms": 271.40179998241365,
      "mean_inter_token_latency_ms": 5.428035999648273,
      "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": 443.99880000855774,
      "ttft_ms": 443.9991999533959,
      "visible_generation_ms": 271.4014000375755
    },
    {
      "content_chunks": 18,
      "decode_tok_s": 165.90962720095877,
      "e2e_ms": 714.9273999966681,
      "effective_prefill_tok_s": 8142.973760339138,
      "generation_ms": 271.2319999700412,
      "mean_inter_token_latency_ms": 6.027377777112027,
      "output_token_source": "usage",
      "output_tokens": 46,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 443.6954000266269,
      "ttft_ms": 443.69590003043413,
      "visible_generation_ms": 271.23149996623397
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260903T122900Z",
  "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": "2026-09-03T12:29:00Z",
  "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": 3761.204100039322
}