{
  "base_url": "http://127.0.0.1:39062/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "32768": 3
  },
  "context_seed": 0,
  "context_tokens": 32768,
  "engine": "vllm-r16-b12x",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-02T11:13:18Z",
  "gpu": "2x-rtx-pro-6000-blackwell-max-q-tp2",
  "manifest": null,
  "max_context_tokens": 650000,
  "max_tokens": 1024,
  "measurement_protocol": "capacity-v4-reasoning",
  "metrics": {
    "content_chunks": 24,
    "content_chunks_s": null,
    "decode_tok_s_p50": 141.64780980194067,
    "decode_tok_s_p95": 147.5812218221054,
    "e2e_p50_ms": 5593.984099999943,
    "e2e_p95_ms": 5664.356400000543,
    "effective_prefill_tok_s_p50": 8793.289381685374,
    "effective_prefill_tok_s_p95": 9417.935029708684,
    "generation_p50_ms": 2880.3834000009374,
    "generation_p95_ms": 3006.193000001076,
    "mean_inter_token_latency_ms_p50": 7.0597632352964155,
    "mean_inter_token_latency_ms_p95": 7.534318295742046,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 1122,
    "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": 302,
    "throughput_tok_s": 67.49607430920781,
    "time_to_first_output_p50_ms": 2658.1633999994665,
    "time_to_first_output_p95_ms": 3479.8940000000584,
    "ttft_p50_ms": 5463.510300000053,
    "ttft_p95_ms": 5463.806499999919,
    "ttft_samples": 3
  },
  "model": "deepseek-v4-flash-0731-r16-b12x-dspark5-maxseq16-tp2-650k",
  "request_timings": [
    {
      "content_chunks": 6,
      "decode_tok_s": 147.5812218221054,
      "e2e_ms": 5593.984099999943,
      "effective_prefill_tok_s": 6716.871261021056,
      "generation_ms": 2114.090099999885,
      "mean_inter_token_latency_ms": 6.775929807691939,
      "output_token_source": "usage",
      "output_tokens": 313,
      "planned_context_tokens": 32768,
      "prompt_tokens": 23374,
      "reasoning_chunks": 80,
      "request_index": 0,
      "time_to_first_output_ms": 3479.8940000000584,
      "ttft_ms": 5463.510300000053,
      "visible_generation_ms": 130.47379999989062
    },
    {
      "content_chunks": 9,
      "decode_tok_s": 132.7260092748061,
      "e2e_ms": 5664.356400000543,
      "effective_prefill_tok_s": 8793.289381685374,
      "generation_ms": 3006.193000001076,
      "mean_inter_token_latency_ms": 7.534318295742046,
      "output_token_source": "usage",
      "output_tokens": 400,
      "planned_context_tokens": 32768,
      "prompt_tokens": 23374,
      "reasoning_chunks": 113,
      "request_index": 1,
      "time_to_first_output_ms": 2658.1633999994665,
      "ttft_ms": 5463.806499999919,
      "visible_generation_ms": 200.5499000006239
    },
    {
      "content_chunks": 9,
      "decode_tok_s": 141.64780980194067,
      "e2e_ms": 5362.243800000215,
      "effective_prefill_tok_s": 9417.935029708684,
      "generation_ms": 2880.3834000009374,
      "mean_inter_token_latency_ms": 7.0597632352964155,
      "output_token_source": "usage",
      "output_tokens": 409,
      "planned_context_tokens": 32768,
      "prompt_tokens": 23374,
      "reasoning_chunks": 109,
      "request_index": 2,
      "time_to_first_output_ms": 2481.8603999992774,
      "ttft_ms": 5142.207900000358,
      "visible_generation_ms": 220.03589999985707
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260802T111301Z",
  "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": "2026-08-02T11:13:01Z",
  "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": 16623.18899999991
}