{
  "base_url": "http://127.0.0.1:8081/v1",
  "completed": 1,
  "concurrency": 1,
  "context_distribution": {
    "57344": 1
  },
  "context_seed": 0,
  "context_tokens": 57344,
  "engine": "vllm",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-03T20:57:28Z",
  "gpu": "rtx-5090",
  "manifest": null,
  "max_context_tokens": 65536,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 18,
    "content_chunks_s": null,
    "decode_tok_s_p50": 127.49171123551552,
    "decode_tok_s_p95": 127.49171123551552,
    "e2e_p50_ms": 9340.320500021335,
    "e2e_p95_ms": 9340.320500021335,
    "effective_prefill_tok_s_p50": 6017.743839734982,
    "effective_prefill_tok_s_p95": 6017.743839734982,
    "generation_p50_ms": 415.7132999971509,
    "generation_p95_ms": 415.7132999971509,
    "mean_inter_token_latency_ms_p50": 7.843647169757564,
    "mean_inter_token_latency_ms_p95": 7.843647169757564,
    "output_token_sources": {
      "usage": 1
    },
    "output_tokens": 54,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 1,
    "prompt_tokens": 53706,
    "prompt_tokens_p50": 53706.0,
    "prompt_tokens_p95": 53706.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 5.780227281129249,
    "time_to_first_output_p50_ms": 8924.607200024184,
    "time_to_first_output_p95_ms": 8924.607200024184,
    "ttft_p50_ms": 8924.60830003256,
    "ttft_p95_ms": 8924.60830003256,
    "ttft_samples": 1
  },
  "model": "qwen38-27b-unsloth-nvfp4-mtp3-64k",
  "request_timings": [
    {
      "content_chunks": 18,
      "decode_tok_s": 127.49171123551552,
      "e2e_ms": 9340.320500021335,
      "effective_prefill_tok_s": 6017.743839734982,
      "generation_ms": 415.7132999971509,
      "mean_inter_token_latency_ms": 7.843647169757564,
      "output_token_source": "usage",
      "output_tokens": 54,
      "planned_context_tokens": 57344,
      "prompt_tokens": 53706,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 8924.607200024184,
      "ttft_ms": 8924.60830003256,
      "visible_generation_ms": 415.71219998877496
    }
  ],
  "requests": 1,
  "run_id": "benchmark-20260903T205718Z",
  "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": "2026-09-03T20:57:18Z",
  "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": 9342.193199961912
}