{
  "base_url": "http://127.0.0.1:8081/v1",
  "completed": 5,
  "concurrency": 1,
  "context_distribution": {
    "4096": 5
  },
  "context_seed": 0,
  "context_tokens": 4096,
  "engine": "vllm",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-03T20:57:14Z",
  "gpu": "rtx-5090",
  "manifest": null,
  "max_context_tokens": 65536,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 72,
    "content_chunks_s": null,
    "decode_tok_s_p50": 137.67061952021479,
    "decode_tok_s_p95": 141.84478599416798,
    "e2e_p50_ms": 698.6680999980308,
    "e2e_p95_ms": 811.961599974893,
    "effective_prefill_tok_s_p50": 9295.002489548891,
    "effective_prefill_tok_s_p95": 9309.681812166184,
    "generation_p50_ms": 308.77960001816973,
    "generation_p95_ms": 403.4739999915473,
    "mean_inter_token_latency_ms_p50": 7.263713953529247,
    "mean_inter_token_latency_ms_p95": 8.966088888701051,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 217,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 5,
    "prompt_tokens": 18065,
    "prompt_tokens_p50": 3613.0,
    "prompt_tokens_p95": 3613.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 60.749266333086936,
    "time_to_first_output_p50_ms": 388.7034999788739,
    "time_to_first_output_p95_ms": 408.4875999833457,
    "ttft_p50_ms": 388.7038999819197,
    "ttft_p95_ms": 408.4883999894373,
    "ttft_samples": 5
  },
  "model": "qwen38-27b-unsloth-nvfp4-mtp3-64k",
  "request_timings": [
    {
      "content_chunks": 18,
      "decode_tok_s": 111.5313502256471,
      "e2e_ms": 811.961599974893,
      "effective_prefill_tok_s": 8844.821728119297,
      "generation_ms": 403.4739999915473,
      "mean_inter_token_latency_ms": 8.966088888701051,
      "output_token_source": "usage",
      "output_tokens": 46,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 408.4875999833457,
      "ttft_ms": 408.4883999894373,
      "visible_generation_ms": 403.4731999854557
    },
    {
      "content_chunks": 14,
      "decode_tok_s": 137.67061952021479,
      "e2e_ms": 700.4302999703214,
      "effective_prefill_tok_s": 9309.681812166184,
      "generation_ms": 312.3397000017576,
      "mean_inter_token_latency_ms": 7.263713953529247,
      "output_token_source": "usage",
      "output_tokens": 44,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 388.09059996856377,
      "ttft_ms": 388.0912999738939,
      "visible_generation_ms": 312.33899999642745
    },
    {
      "content_chunks": 13,
      "decode_tok_s": 141.84478599416798,
      "e2e_ms": 684.7108000074513,
      "effective_prefill_tok_s": 9297.179066922963,
      "generation_ms": 296.0983000230044,
      "mean_inter_token_latency_ms": 7.049959524357248,
      "output_token_source": "usage",
      "output_tokens": 43,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 388.6124999844469,
      "ttft_ms": 388.6128999874927,
      "visible_generation_ms": 296.0979000199586
    },
    {
      "content_chunks": 13,
      "decode_tok_s": 139.78159823050706,
      "e2e_ms": 674.8642000020482,
      "effective_prefill_tok_s": 9295.002489548891,
      "generation_ms": 286.1607000231743,
      "mean_inter_token_latency_ms": 7.154017500579357,
      "output_token_source": "usage",
      "output_tokens": 41,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 388.7034999788739,
      "ttft_ms": 388.7038999819197,
      "visible_generation_ms": 286.1603000201285
    },
    {
      "content_chunks": 14,
      "decode_tok_s": 136.01934842045446,
      "e2e_ms": 698.6680999980308,
      "effective_prefill_tok_s": 9266.751905189874,
      "generation_ms": 308.77960001816973,
      "mean_inter_token_latency_ms": 7.351895238527851,
      "output_token_source": "usage",
      "output_tokens": 43,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 389.8884999798611,
      "ttft_ms": 389.8889999836683,
      "visible_generation_ms": 308.7791000143625
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260903T205711Z",
  "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": "2026-09-03T20:57:11Z",
  "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": 3572.0596000319347
}