{
  "base_url": "http://127.0.0.1:39064/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "32768": 3
  },
  "context_seed": 0,
  "context_tokens": 32768,
  "engine": "vllm-r16-b12x",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-02T11:18:59Z",
  "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": 21,
    "content_chunks_s": null,
    "decode_tok_s_p50": 119.88018252398287,
    "decode_tok_s_p95": 138.08343288236128,
    "e2e_p50_ms": 5127.170900001147,
    "e2e_p95_ms": 5323.9401999999245,
    "effective_prefill_tok_s_p50": 7901.056089351128,
    "effective_prefill_tok_s_p95": 8961.767674353754,
    "generation_p50_ms": 2168.8322000009066,
    "generation_p95_ms": 2715.749399998458,
    "mean_inter_token_latency_ms_p50": 8.341662307695795,
    "mean_inter_token_latency_ms_p95": 8.609873571419614,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 778,
    "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": 233,
    "throughput_tok_s": 50.486647806639155,
    "time_to_first_output_p50_ms": 2958.3387000002404,
    "time_to_first_output_p95_ms": 3750.630100001217,
    "ttft_p50_ms": 5003.2945000002655,
    "ttft_p95_ms": 5103.858800001035,
    "ttft_samples": 3
  },
  "model": "deepseek-v4-flash-0731-r16-b12x-dspark5-maxseq4-tp2-1m",
  "request_timings": [
    {
      "content_chunks": 5,
      "decode_tok_s": 116.14572405795708,
      "e2e_ms": 4956.012399999963,
      "effective_prefill_tok_s": 6232.019521197895,
      "generation_ms": 1205.382299998746,
      "mean_inter_token_latency_ms": 8.609873571419614,
      "output_token_source": "usage",
      "output_tokens": 141,
      "planned_context_tokens": 32768,
      "prompt_tokens": 23374,
      "reasoning_chunks": 46,
      "request_index": 0,
      "time_to_first_output_ms": 3750.630100001217,
      "ttft_ms": 4854.434599999877,
      "visible_generation_ms": 101.57780000008643
    },
    {
      "content_chunks": 6,
      "decode_tok_s": 119.88018252398287,
      "e2e_ms": 5127.170900001147,
      "effective_prefill_tok_s": 7901.056089351128,
      "generation_ms": 2168.8322000009066,
      "mean_inter_token_latency_ms": 8.341662307695795,
      "output_token_source": "usage",
      "output_tokens": 261,
      "planned_context_tokens": 32768,
      "prompt_tokens": 23374,
      "reasoning_chunks": 84,
      "request_index": 1,
      "time_to_first_output_ms": 2958.3387000002404,
      "ttft_ms": 5003.2945000002655,
      "visible_generation_ms": 123.8764000008814
    },
    {
      "content_chunks": 10,
      "decode_tok_s": 138.08343288236128,
      "e2e_ms": 5323.9401999999245,
      "effective_prefill_tok_s": 8961.767674353754,
      "generation_ms": 2715.749399998458,
      "mean_inter_token_latency_ms": 7.241998399995889,
      "output_token_source": "usage",
      "output_tokens": 376,
      "planned_context_tokens": 32768,
      "prompt_tokens": 23374,
      "reasoning_chunks": 103,
      "request_index": 2,
      "time_to_first_output_ms": 2608.1908000014664,
      "ttft_ms": 5103.858800001035,
      "visible_generation_ms": 220.08139999888954
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260802T111844Z",
  "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": "2026-08-02T11:18:44Z",
  "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": 15410.014999999476
}