{
  "base_url": "http://127.0.0.1:30010/v1",
  "completed": 10,
  "concurrency": 1,
  "context_distribution": {
    "8192": 10
  },
  "context_seed": 0,
  "context_tokens": 8192,
  "engine": "vllm-7b9cb5b7-ngc-26.06",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-07-29T10:32:34Z",
  "gpu": "rtx-pro-6000",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 64,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 503,
    "content_chunks_s": null,
    "decode_tok_s_p50": 78.73069739797482,
    "decode_tok_s_p95": 78.89059016548983,
    "e2e_p50_ms": 748.0625999996846,
    "e2e_p95_ms": 1671.8797000030463,
    "effective_prefill_tok_s_p50": 50214.904216362556,
    "effective_prefill_tok_s_p95": 53178.64583932795,
    "generation_p50_ms": 609.7364000015659,
    "generation_p95_ms": 774.4170000005397,
    "mean_inter_token_latency_ms_p50": 12.695758139459432,
    "mean_inter_token_latency_ms_p95": 12.712432758642786,
    "output_token_sources": {
      "usage": 10
    },
    "output_tokens": 513,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 10,
    "prompt_tokens": 73525,
    "prompt_tokens_p50": 7352.0,
    "prompt_tokens_p95": 7356.0,
    "throughput_tok_s": 59.454226197024134,
    "ttft_p50_ms": 145.81790000011097,
    "ttft_p95_ms": 934.5586000017647,
    "ttft_samples": 10
  },
  "model": "qwen35-122b-a10b-nvfp4-video-262k",
  "request_timings": [
    {
      "decode_tok_s": 78.66314960998564,
      "e2e_ms": 1671.8797000030463,
      "effective_prefill_tok_s": 7864.675366516472,
      "generation_ms": 737.3211000012816,
      "mean_inter_token_latency_ms": 12.712432758642786,
      "output_token_source": "usage",
      "output_tokens": 59,
      "planned_context_tokens": 8192,
      "prompt_tokens": 7350,
      "request_index": 0,
      "ttft_ms": 934.5586000017647
    },
    {
      "decode_tok_s": 78.88672054708977,
      "e2e_ms": 677.495000003546,
      "effective_prefill_tok_s": 50659.60878359983,
      "generation_ms": 532.4090000030992,
      "mean_inter_token_latency_ms": 12.676404761978553,
      "output_token_source": "usage",
      "output_tokens": 43,
      "planned_context_tokens": 8192,
      "prompt_tokens": 7350,
      "request_index": 1,
      "ttft_ms": 145.0860000004468
    },
    {
      "decode_tok_s": 78.73069739797482,
      "e2e_ms": 778.6285000001953,
      "effective_prefill_tok_s": 51200.88720366849,
      "generation_ms": 635.0763000009465,
      "mean_inter_token_latency_ms": 12.70152600001893,
      "output_token_source": "usage",
      "output_tokens": 51,
      "planned_context_tokens": 8192,
      "prompt_tokens": 7350,
      "request_index": 2,
      "ttft_ms": 143.55219999924884
    },
    {
      "decode_tok_s": 78.6979849009924,
      "e2e_ms": 819.8514999967301,
      "effective_prefill_tok_s": 50214.904216362556,
      "generation_ms": 673.4606999998505,
      "mean_inter_token_latency_ms": 12.706805660374537,
      "output_token_source": "usage",
      "output_tokens": 54,
      "planned_context_tokens": 8192,
      "prompt_tokens": 7351,
      "request_index": 3,
      "ttft_ms": 146.39079999687965
    },
    {
      "decode_tok_s": 78.76893198361799,
      "e2e_ms": 921.769000000495,
      "effective_prefill_tok_s": 49894.13106033329,
      "generation_ms": 774.4170000005397,
      "mean_inter_token_latency_ms": 12.695360655746553,
      "output_token_source": "usage",
      "output_tokens": 62,
      "planned_context_tokens": 8192,
      "prompt_tokens": 7352,
      "request_index": 4,
      "ttft_ms": 147.3519999999553
    },
    {
      "decode_tok_s": 78.8359519449426,
      "e2e_ms": 741.9926000002306,
      "effective_prefill_tok_s": 50425.907930332316,
      "generation_ms": 596.1747000001196,
      "mean_inter_token_latency_ms": 12.684568085108928,
      "output_token_source": "usage",
      "output_tokens": 48,
      "planned_context_tokens": 8192,
      "prompt_tokens": 7353,
      "request_index": 5,
      "ttft_ms": 145.81790000011097
    },
    {
      "decode_tok_s": 78.6707785261556,
      "e2e_ms": 831.6161000002467,
      "effective_prefill_tok_s": 50636.55514363892,
      "generation_ms": 686.4047999988543,
      "mean_inter_token_latency_ms": 12.711199999978783,
      "output_token_source": "usage",
      "output_tokens": 55,
      "planned_context_tokens": 8192,
      "prompt_tokens": 7353,
      "request_index": 6,
      "ttft_ms": 145.21130000139237
    },
    {
      "decode_tok_s": 78.89059016548983,
      "e2e_ms": 742.3608999997668,
      "effective_prefill_tok_s": 50164.01874211392,
      "generation_ms": 595.761799999309,
      "mean_inter_token_latency_ms": 12.675782978708702,
      "output_token_source": "usage",
      "output_tokens": 48,
      "planned_context_tokens": 8192,
      "prompt_tokens": 7354,
      "request_index": 7,
      "ttft_ms": 146.59910000045784
    },
    {
      "decode_tok_s": 78.72254305282861,
      "e2e_ms": 748.0625999996846,
      "effective_prefill_tok_s": 53178.64583932795,
      "generation_ms": 609.7364000015659,
      "mean_inter_token_latency_ms": 12.70284166669929,
      "output_token_source": "usage",
      "output_tokens": 49,
      "planned_context_tokens": 8192,
      "prompt_tokens": 7356,
      "request_index": 8,
      "ttft_ms": 138.32619999811868
    },
    {
      "decode_tok_s": 78.76646585538835,
      "e2e_ms": 692.7184999985911,
      "effective_prefill_tok_s": 50108.6846191544,
      "generation_ms": 545.9175999967556,
      "mean_inter_token_latency_ms": 12.695758139459432,
      "output_token_source": "usage",
      "output_tokens": 44,
      "planned_context_tokens": 8192,
      "prompt_tokens": 7356,
      "request_index": 9,
      "ttft_ms": 146.80090000183554
    }
  ],
  "requests": 10,
  "run_id": "benchmark-20260729T103225Z",
  "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": "2026-07-29T10:32:25Z",
  "tier": null,
  "timing_methodology": {
    "clock": "client time.perf_counter",
    "decode": "usage completion tokens after the first token divided by client-observed generation time",
    "effective_prefill": "usage.prompt_tokens divided by client-observed TTFT; includes queueing, scheduling, prefill, and first-token work",
    "generation": "client-observed E2E minus TTFT",
    "mean_inter_token_latency": "client-observed generation time divided by completion tokens after the first token; not a raw per-token timestamp trace",
    "ttft": "request start through the first non-empty streamed content delta"
  },
  "wall_clock_ms": 8628.486700003123
}