{
  "base_url": "http://127.0.0.1:39268/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "32768": 3
  },
  "context_seed": 0,
  "context_tokens": 32768,
  "engine": "sglang-5f55db35-dflash2-k12",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-05T03:14:27Z",
  "gpu": "RTX PRO 6000 Blackwell Max-Q TP1",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 512,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 55,
    "content_chunks_s": null,
    "decode_tok_s_p50": 85.2288412874749,
    "decode_tok_s_p95": 85.4436125631858,
    "e2e_p50_ms": 4300.392900011502,
    "e2e_p95_ms": 4611.668400000781,
    "effective_prefill_tok_s_p50": 7922.791407253229,
    "effective_prefill_tok_s_p95": 8209.200751660786,
    "generation_p50_ms": 608.5884999483824,
    "generation_p95_ms": 650.0372000737116,
    "mean_inter_token_latency_ms_p50": 11.733117391881736,
    "mean_inter_token_latency_ms_p95": 12.745827452425717,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 152,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 3,
    "prompt_tokens": 89384,
    "prompt_tokens_p50": 29795.0,
    "prompt_tokens_p95": 29797.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 11.520429093532137,
    "time_to_first_output_p50_ms": 3760.6694999849424,
    "time_to_first_output_p95_ms": 4003.0799000523984,
    "ttft_p50_ms": 3760.670099989511,
    "ttft_p95_ms": 4003.0805999413133,
    "ttft_samples": 3
  },
  "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12",
  "request_timings": [
    {
      "content_chunks": 18,
      "decode_tok_s": 85.4436125631858,
      "e2e_ms": 4611.668400000781,
      "effective_prefill_tok_s": 7442.26963833773,
      "generation_ms": 608.5884999483824,
      "mean_inter_token_latency_ms": 11.703624999007353,
      "output_token_source": "usage",
      "output_tokens": 53,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29792,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 4003.0799000523984,
      "ttft_ms": 4003.0805999413133,
      "visible_generation_ms": 608.5878000594676
    },
    {
      "content_chunks": 17,
      "decode_tok_s": 85.2288412874749,
      "e2e_ms": 4300.392900011502,
      "effective_prefill_tok_s": 7922.791407253229,
      "generation_ms": 539.7234000265598,
      "mean_inter_token_latency_ms": 11.733117391881736,
      "output_token_source": "usage",
      "output_tokens": 47,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29795,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 3760.6694999849424,
      "ttft_ms": 3760.670099989511,
      "visible_generation_ms": 539.7228000219911
    },
    {
      "content_chunks": 20,
      "decode_tok_s": 78.45704829541572,
      "e2e_ms": 4279.745000065304,
      "effective_prefill_tok_s": 8209.200751660786,
      "generation_ms": 650.0372000737116,
      "mean_inter_token_latency_ms": 12.745827452425717,
      "output_token_source": "usage",
      "output_tokens": 52,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29797,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 3629.7077999915928,
      "ttft_ms": 3629.7081999946386,
      "visible_generation_ms": 650.0368000706658
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260905T031414Z",
  "schema": "anvil-serving.benchmark/v1",
  "serve_flags": {
    "no_thinking": true,
    "reasoning_effort": null,
    "shared_prefix_burst": false,
    "thinking_mode": "disabled"
  },
  "source_recipe": null,
  "started_at": "2026-09-05T03:14:14Z",
  "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": 13193.95300000906
}