{
  "base_url": "http://127.0.0.1:39081/v1",
  "completed": 5,
  "concurrency": 1,
  "context_distribution": {
    "131072": 5
  },
  "context_seed": 38,
  "context_tokens": 131072,
  "engine": "sglang-d91c3682b-sm120-qsa-pr36556-mtp3",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-27T03:43:52Z",
  "gpu": "2x-rtx-pro-6000-blackwell-max-q-tp2-pcie-wsl2",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 512,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 65,
    "content_chunks_s": null,
    "decode_tok_s_p50": 114.69016452454629,
    "decode_tok_s_p95": 124.2016773839353,
    "e2e_p50_ms": 7192.358999978751,
    "e2e_p95_ms": 9057.573900092393,
    "effective_prefill_tok_s_p50": 18095.841242788025,
    "effective_prefill_tok_s_p95": 18186.027414996366,
    "generation_p50_ms": 210.7321999501437,
    "generation_p95_ms": 305.95399998128414,
    "mean_inter_token_latency_ms_p50": 8.719143477956885,
    "mean_inter_token_latency_ms_p95": 9.162269563049726,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 146,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 5,
    "prompt_tokens": 627233,
    "prompt_tokens_p50": 125447.0,
    "prompt_tokens_p95": 125450.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 3.864569405070692,
    "time_to_first_output_p50_ms": 6932.366299908608,
    "time_to_first_output_p95_ms": 8846.84170014225,
    "ttft_p50_ms": 6932.366699911654,
    "ttft_p95_ms": 8846.842200029641,
    "ttft_samples": 5
  },
  "model": "qwen38-flash-next-radixark-nvfp4-sglang-qsa-fast-tp2-262k-mtp3",
  "request_timings": [
    {
      "content_chunks": 10,
      "decode_tok_s": 114.69016452454629,
      "e2e_ms": 7172.919900156558,
      "effective_prefill_tok_s": 17991.275173408165,
      "generation_ms": 200.54029999300838,
      "mean_inter_token_latency_ms": 8.719143477956885,
      "output_token_source": "usage",
      "output_tokens": 24,
      "planned_context_tokens": 131072,
      "prompt_tokens": 125442,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 6972.379600163549,
      "ttft_ms": 6972.380300052464,
      "visible_generation_ms": 200.53960010409355
    },
    {
      "content_chunks": 11,
      "decode_tok_s": 109.2644885751474,
      "e2e_ms": 7108.321499777958,
      "effective_prefill_tok_s": 18186.027414996366,
      "generation_ms": 210.49839979968965,
      "mean_inter_token_latency_ms": 9.152104339116942,
      "output_token_source": "usage",
      "output_tokens": 24,
      "planned_context_tokens": 131072,
      "prompt_tokens": 125444,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 6897.823099978268,
      "ttft_ms": 6897.823999868706,
      "visible_generation_ms": 210.49749990925193
    },
    {
      "content_chunks": 17,
      "decode_tok_s": 124.2016773839353,
      "e2e_ms": 7238.320299889892,
      "effective_prefill_tok_s": 18095.841242788025,
      "generation_ms": 305.95399998128414,
      "mean_inter_token_latency_ms": 8.051421052139057,
      "output_token_source": "usage",
      "output_tokens": 39,
      "planned_context_tokens": 131072,
      "prompt_tokens": 125447,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 6932.366299908608,
      "ttft_ms": 6932.366699911654,
      "visible_generation_ms": 305.95359997823834
    },
    {
      "content_chunks": 16,
      "decode_tok_s": 116.3777361136696,
      "e2e_ms": 7192.358999978751,
      "effective_prefill_tok_s": 18180.61426543098,
      "generation_ms": 292.1521000098437,
      "mean_inter_token_latency_ms": 8.592708823818933,
      "output_token_source": "usage",
      "output_tokens": 35,
      "planned_context_tokens": 131072,
      "prompt_tokens": 125450,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 6900.206899968907,
      "ttft_ms": 6900.207199854776,
      "visible_generation_ms": 292.1518001239747
    },
    {
      "content_chunks": 11,
      "decode_tok_s": 109.14326337143294,
      "e2e_ms": 9057.573900092393,
      "effective_prefill_tok_s": 14180.201732103205,
      "generation_ms": 210.7321999501437,
      "mean_inter_token_latency_ms": 9.162269563049726,
      "output_token_source": "usage",
      "output_tokens": 24,
      "planned_context_tokens": 131072,
      "prompt_tokens": 125450,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 8846.84170014225,
      "ttft_ms": 8846.842200029641,
      "visible_generation_ms": 210.73170006275177
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260827T034315Z",
  "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": "2026-08-27T03:43:15Z",
  "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": 37779.11190013401
}