{
  "base_url": "http://127.0.0.1:39081/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "196608": 3
  },
  "context_seed": 38,
  "context_tokens": 196608,
  "engine": "sglang-d91c3682b-sm120-qsa-pr36556-mtp3",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-27T03:44:54Z",
  "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": 36,
    "content_chunks_s": null,
    "decode_tok_s_p50": 120.46728737700286,
    "decode_tok_s_p95": 138.26033244594365,
    "e2e_p50_ms": 20256.702600046992,
    "e2e_p95_ms": 20592.20950002782,
    "effective_prefill_tok_s_p50": 9429.43686500671,
    "effective_prefill_tok_s_p95": 9566.33064172278,
    "generation_p50_ms": 192.67629995010793,
    "generation_p95_ms": 289.3093000166118,
    "mean_inter_token_latency_ms_p50": 8.301008695169635,
    "mean_inter_token_latency_ms_p95": 8.377230432613388,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 89,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 3,
    "prompt_tokens": 567627,
    "prompt_tokens_p50": 189209.0,
    "prompt_tokens_p95": 189210.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 1.4607813789510204,
    "time_to_first_output_p50_ms": 20065.77940005809,
    "time_to_first_output_p95_ms": 20399.533200077713,
    "ttft_p50_ms": 20065.780100179836,
    "ttft_p95_ms": 20399.533899966627,
    "ttft_samples": 3
  },
  "model": "qwen38-flash-next-radixark-nvfp4-sglang-qsa-fast-tp2-262k-mtp3",
  "request_timings": [
    {
      "content_chunks": 10,
      "decode_tok_s": 119.37119410096454,
      "e2e_ms": 20592.20950002782,
      "effective_prefill_tok_s": 9275.11419718561,
      "generation_ms": 192.67629995010793,
      "mean_inter_token_latency_ms": 8.377230432613388,
      "output_token_source": "usage",
      "output_tokens": 24,
      "planned_context_tokens": 196608,
      "prompt_tokens": 189208,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 20399.533200077713,
      "ttft_ms": 20399.533899966627,
      "visible_generation_ms": 192.6756000611931
    },
    {
      "content_chunks": 10,
      "decode_tok_s": 120.46728737700286,
      "e2e_ms": 20256.702600046992,
      "effective_prefill_tok_s": 9429.43686500671,
      "generation_ms": 190.92319998890162,
      "mean_inter_token_latency_ms": 8.301008695169635,
      "output_token_source": "usage",
      "output_tokens": 24,
      "planned_context_tokens": 196608,
      "prompt_tokens": 189209,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 20065.77940005809,
      "ttft_ms": 20065.780100179836,
      "visible_generation_ms": 190.92249986715615
    },
    {
      "content_chunks": 16,
      "decode_tok_s": 138.26033244594365,
      "e2e_ms": 20068.052800139412,
      "effective_prefill_tok_s": 9566.33064172278,
      "generation_ms": 289.3093000166118,
      "mean_inter_token_latency_ms": 7.232732500415295,
      "output_token_source": "usage",
      "output_tokens": 41,
      "planned_context_tokens": 196608,
      "prompt_tokens": 189210,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 19778.7435001228,
      "ttft_ms": 19778.74410012737,
      "visible_generation_ms": 289.3087000120431
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260827T034353Z",
  "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": "2026-08-27T03:43:53Z",
  "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": 60926.2969000265
}