{
  "base_url": "http://127.0.0.1:39081/v1",
  "completed": 5,
  "concurrency": 1,
  "context_distribution": {
    "32768": 5
  },
  "context_seed": 38,
  "context_tokens": 32768,
  "engine": "sglang-d91c3682b-sm120-qsa-pr36556-mtp3",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-27T03:42:56Z",
  "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": 83,
    "content_chunks_s": null,
    "decode_tok_s_p50": 130.09063380992774,
    "decode_tok_s_p95": 156.51773372354495,
    "e2e_p50_ms": 2569.3661000113934,
    "e2e_p95_ms": 2925.5693000741303,
    "effective_prefill_tok_s_p50": 13260.964123876489,
    "effective_prefill_tok_s_p95": 13319.420141804705,
    "generation_p50_ms": 329.24539991654456,
    "generation_p95_ms": 359.40740001387894,
    "mean_inter_token_latency_ms_p50": 7.686948481326302,
    "mean_inter_token_latency_ms_p95": 8.44218974144986,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 209,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 5,
    "prompt_tokens": 148985,
    "prompt_tokens_p50": 29797.0,
    "prompt_tokens_p95": 29801.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 15.938849604323185,
    "time_to_first_output_p50_ms": 2247.272500069812,
    "time_to_first_output_p95_ms": 2596.3239001575857,
    "ttft_p50_ms": 2247.272799955681,
    "ttft_p95_ms": 2596.325300168246,
    "ttft_samples": 5
  },
  "model": "qwen38-flash-next-radixark-nvfp4-sglang-qsa-fast-tp2-262k-mtp3",
  "request_timings": [
    {
      "content_chunks": 17,
      "decode_tok_s": 118.45268000672301,
      "e2e_ms": 2925.5693000741303,
      "effective_prefill_tok_s": 11474.685418946285,
      "generation_ms": 329.24539991654456,
      "mean_inter_token_latency_ms": 8.44218974144986,
      "output_token_source": "usage",
      "output_tokens": 40,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29792,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 2596.3239001575857,
      "ttft_ms": 2596.325300168246,
      "visible_generation_ms": 329.24399990588427
    },
    {
      "content_chunks": 18,
      "decode_tok_s": 132.51686881235327,
      "e2e_ms": 2569.3661000113934,
      "effective_prefill_tok_s": 13317.195665996262,
      "generation_ms": 332.0331999566406,
      "mean_inter_token_latency_ms": 7.54620908992365,
      "output_token_source": "usage",
      "output_tokens": 45,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29795,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 2237.332900054753,
      "ttft_ms": 2237.3334001749754,
      "visible_generation_ms": 332.03269983641803
    },
    {
      "content_chunks": 15,
      "decode_tok_s": 156.51773372354495,
      "e2e_ms": 2505.4496000520885,
      "effective_prefill_tok_s": 13319.420141804705,
      "generation_ms": 268.34020018577576,
      "mean_inter_token_latency_ms": 6.389052385375614,
      "output_token_source": "usage",
      "output_tokens": 43,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29797,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 2237.1093998663127,
      "ttft_ms": 2237.1096999850124,
      "visible_generation_ms": 268.3399000670761
    },
    {
      "content_chunks": 14,
      "decode_tok_s": 130.09063380992774,
      "e2e_ms": 2501.707299845293,
      "effective_prefill_tok_s": 13256.003679879976,
      "generation_ms": 253.66929988376796,
      "mean_inter_token_latency_ms": 7.686948481326302,
      "output_token_source": "usage",
      "output_tokens": 34,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29800,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 2248.037999961525,
      "ttft_ms": 2248.038499848917,
      "visible_generation_ms": 253.66879999637604
    },
    {
      "content_chunks": 19,
      "decode_tok_s": 127.98846100059058,
      "e2e_ms": 2606.679900083691,
      "effective_prefill_tok_s": 13260.964123876489,
      "generation_ms": 359.40740001387894,
      "mean_inter_token_latency_ms": 7.813204348127803,
      "output_token_source": "usage",
      "output_tokens": 47,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29801,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 2247.272500069812,
      "ttft_ms": 2247.272799955681,
      "visible_generation_ms": 359.4071001280099
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260827T034242Z",
  "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": "2026-08-27T03:42:42Z",
  "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": 13112.615100108087
}