{
  "base_url": "http://127.0.0.1:39065/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "32768": 3
  },
  "context_seed": 0,
  "context_tokens": 32768,
  "engine": "vllm-r16-b12x",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-02T11:32:34Z",
  "gpu": "2x-rtx-pro-6000-blackwell-max-q-tp2",
  "manifest": null,
  "max_context_tokens": 1000000,
  "max_tokens": 1024,
  "measurement_protocol": "capacity-v4-reasoning",
  "metrics": {
    "content_chunks": 31,
    "content_chunks_s": null,
    "decode_tok_s_p50": 129.01234990465952,
    "decode_tok_s_p95": 138.5399992736546,
    "e2e_p50_ms": 4807.090199999948,
    "e2e_p95_ms": 6843.9619000000675,
    "effective_prefill_tok_s_p50": 8012.208282727796,
    "effective_prefill_tok_s_p95": 9096.146077239307,
    "generation_p50_ms": 1139.4258000000264,
    "generation_p95_ms": 3926.6638000008243,
    "mean_inter_token_latency_ms_p50": 7.751195918367526,
    "mean_inter_token_latency_ms_p95": 7.79814074074373,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 775,
    "prefix_cache_hit_avg": 0.1387296426228573,
    "prompt_token_samples": 3,
    "prompt_tokens": 70122,
    "prompt_tokens_p50": 23374.0,
    "prompt_tokens_p95": 23374.0,
    "reasoning_chunks": 202,
    "throughput_tok_s": 52.170292937338594,
    "time_to_first_output_p50_ms": 2917.2980999992433,
    "time_to_first_output_p95_ms": 3667.6643999999214,
    "ttft_p50_ms": 4579.575499999919,
    "ttft_p95_ms": 6643.280599999343,
    "ttft_samples": 3
  },
  "model": "deepseek-v4-flash-0731-r16-b12x-dspark5-maxseq16-tp2-1m",
  "request_timings": [
    {
      "content_chunks": 10,
      "decode_tok_s": 129.01234990465952,
      "e2e_ms": 4807.090199999948,
      "effective_prefill_tok_s": 6372.993123362241,
      "generation_ms": 1139.4258000000264,
      "mean_inter_token_latency_ms": 7.751195918367526,
      "output_token_source": "usage",
      "output_tokens": 148,
      "planned_context_tokens": 32768,
      "prompt_tokens": 23374,
      "reasoning_chunks": 38,
      "request_index": 0,
      "time_to_first_output_ms": 3667.6643999999214,
      "ttft_ms": 4579.575499999919,
      "visible_generation_ms": 227.51470000002882
    },
    {
      "content_chunks": 8,
      "decode_tok_s": 138.5399992736546,
      "e2e_ms": 6843.9619000000675,
      "effective_prefill_tok_s": 8012.208282727796,
      "generation_ms": 3926.6638000008243,
      "mean_inter_token_latency_ms": 7.218131985295633,
      "output_token_source": "usage",
      "output_tokens": 545,
      "planned_context_tokens": 32768,
      "prompt_tokens": 23374,
      "reasoning_chunks": 150,
      "request_index": 1,
      "time_to_first_output_ms": 2917.2980999992433,
      "ttft_ms": 6643.280599999343,
      "visible_generation_ms": 200.68130000072415
    },
    {
      "content_chunks": 13,
      "decode_tok_s": 128.2356953081392,
      "e2e_ms": 3201.309100000799,
      "effective_prefill_tok_s": 9096.146077239307,
      "generation_ms": 631.6494000002422,
      "mean_inter_token_latency_ms": 7.79814074074373,
      "output_token_source": "usage",
      "output_tokens": 82,
      "planned_context_tokens": 32768,
      "prompt_tokens": 23374,
      "reasoning_chunks": 14,
      "request_index": 2,
      "time_to_first_output_ms": 2569.6597000005568,
      "ttft_ms": 2907.092500001454,
      "visible_generation_ms": 294.2165999993449
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260802T113220Z",
  "schema": "anvil-serving.benchmark/v1",
  "serve_flags": {
    "no_thinking": false,
    "reasoning_effort": "low",
    "shared_prefix_burst": false,
    "thinking_mode": "default"
  },
  "source_recipe": null,
  "started_at": "2026-08-02T11:32:20Z",
  "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": 14855.19740000018
}