{
  "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": "2026-09-03T12:28:56Z",
  "gpu": "rtx-5090",
  "manifest": null,
  "max_context_tokens": 252928,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 92,
    "content_chunks_s": null,
    "decode_tok_s_p50": 165.8621737067472,
    "decode_tok_s_p95": 184.27738405475665,
    "e2e_p50_ms": 719.8391999700107,
    "e2e_p95_ms": 919.3350999848917,
    "effective_prefill_tok_s_p50": 8410.141145569029,
    "effective_prefill_tok_s_p95": 8435.581389563522,
    "generation_p50_ms": 271.74409996951,
    "generation_p95_ms": 303.2107999897562,
    "mean_inter_token_latency_ms_p50": 6.029102221752207,
    "mean_inter_token_latency_ms_p95": 6.089978571426833,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 241,
    "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": 64.1359831877563,
    "time_to_first_output_p50_ms": 429.6003999770619,
    "time_to_first_output_p95_ms": 647.5910000153817,
    "ttft_p50_ms": 429.6007999801077,
    "ttft_p95_ms": 647.59209996555,
    "ttft_samples": 5
  },
  "model": "qwen38-27b-ninfer-nvfp4-mtp3-252k",
  "request_timings": [
    {
      "content_chunks": 18,
      "decode_tok_s": 165.59697158116424,
      "e2e_ms": 919.3350999848917,
      "effective_prefill_tok_s": 5579.1386846237565,
      "generation_ms": 271.74409996951,
      "mean_inter_token_latency_ms": 6.038757777100222,
      "output_token_source": "usage",
      "output_tokens": 46,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 647.5910000153817,
      "ttft_ms": 647.59209996555,
      "visible_generation_ms": 271.7430000193417
    },
    {
      "content_chunks": 18,
      "decode_tok_s": 165.8621737067472,
      "e2e_ms": 699.6143999858759,
      "effective_prefill_tok_s": 8435.581389563522,
      "generation_ms": 271.3095999788493,
      "mean_inter_token_latency_ms": 6.029102221752207,
      "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": 428.30480000702664,
      "ttft_ms": 428.3057000138797,
      "visible_generation_ms": 271.30869997199625
    },
    {
      "content_chunks": 20,
      "decode_tok_s": 168.19981346879138,
      "e2e_ms": 732.0406999788247,
      "effective_prefill_tok_s": 8425.252064028418,
      "generation_ms": 303.2107999897562,
      "mean_inter_token_latency_ms": 5.9453098037207095,
      "output_token_source": "usage",
      "output_tokens": 52,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 428.8298999890685,
      "ttft_ms": 428.83039999287575,
      "visible_generation_ms": 303.21029998594895
    },
    {
      "content_chunks": 17,
      "decode_tok_s": 164.20419025640481,
      "e2e_ms": 685.3794999769889,
      "effective_prefill_tok_s": 8410.141145569029,
      "generation_ms": 255.77909999992698,
      "mean_inter_token_latency_ms": 6.089978571426833,
      "output_token_source": "usage",
      "output_tokens": 43,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 429.6003999770619,
      "ttft_ms": 429.6007999801077,
      "visible_generation_ms": 255.7786999968812
    },
    {
      "content_chunks": 19,
      "decode_tok_s": 184.27738405475665,
      "e2e_ms": 719.8391999700107,
      "effective_prefill_tok_s": 8358.989083431718,
      "generation_ms": 287.6098999986425,
      "mean_inter_token_latency_ms": 5.426601886766839,
      "output_token_source": "usage",
      "output_tokens": 54,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 432.22929997136816,
      "ttft_ms": 432.22969997441396,
      "visible_generation_ms": 287.6094999955967
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260903T122852Z",
  "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": "2026-09-03T12:28:52Z",
  "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": 3757.640999974683
}