{
  "base_url": "http://127.0.0.1:8081/v1",
  "completed": 5,
  "concurrency": 1,
  "context_distribution": {
    "4096": 5
  },
  "context_seed": 38027,
  "context_tokens": 4096,
  "engine": "ninfer-e3aeaf8c",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-03T12:24:18Z",
  "gpu": "rtx-5090",
  "manifest": null,
  "max_context_tokens": 252928,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 226,
    "content_chunks_s": null,
    "decode_tok_s_p50": 75.30821140384857,
    "decode_tok_s_p95": 75.33176429562124,
    "e2e_p50_ms": 1034.399800002575,
    "e2e_p95_ms": 1182.9052999964915,
    "effective_prefill_tok_s_p50": 8288.330201345752,
    "effective_prefill_tok_s_p95": 8320.835575199959,
    "generation_p50_ms": 597.5444000214338,
    "generation_p95_ms": 637.9563999944367,
    "mean_inter_token_latency_ms_p50": 13.278764444920752,
    "mean_inter_token_latency_ms_p95": 13.29075833321743,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 231,
    "prefix_cache_hit_avg": 0.0,
    "prompt_token_samples": 5,
    "prompt_tokens": 18065,
    "prompt_tokens_p50": 3613.0,
    "prompt_tokens_p95": 3613.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 43.54537631792416,
    "time_to_first_output_p50_ms": 435.9140999731608,
    "time_to_first_output_p95_ms": 558.9984999969602,
    "ttft_p50_ms": 435.9143999754451,
    "ttft_p95_ms": 558.9993000030518,
    "ttft_samples": 5
  },
  "model": "qwen38-27b-ninfer-nvfp4-nospec-252k",
  "request_timings": [
    {
      "content_chunks": 47,
      "decode_tok_s": 75.33176429562124,
      "e2e_ms": 1182.9052999964915,
      "effective_prefill_tok_s": 6463.3447138402835,
      "generation_ms": 623.9067999995314,
      "mean_inter_token_latency_ms": 13.274612765947476,
      "output_token_source": "usage",
      "output_tokens": 48,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 558.9984999969602,
      "ttft_ms": 558.9993000030518,
      "visible_generation_ms": 623.9059999934398
    },
    {
      "content_chunks": 45,
      "decode_tok_s": 75.32148465305177,
      "e2e_ms": 1031.6502999630757,
      "effective_prefill_tok_s": 8320.835575199959,
      "generation_ms": 597.4390999763273,
      "mean_inter_token_latency_ms": 13.276424443918383,
      "output_token_source": "usage",
      "output_tokens": 46,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 434.2111999867484,
      "ttft_ms": 434.2117999913171,
      "visible_generation_ms": 597.4384999717586
    },
    {
      "content_chunks": 41,
      "decode_tok_s": 75.24780386336126,
      "e2e_ms": 980.5397000163794,
      "effective_prefill_tok_s": 8292.91122495103,
      "generation_ms": 544.8664000141434,
      "mean_inter_token_latency_ms": 13.289424390588865,
      "output_token_source": "usage",
      "output_tokens": 42,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 435.6733000022359,
      "ttft_ms": 435.67360000452027,
      "visible_generation_ms": 544.8661000118591
    },
    {
      "content_chunks": 48,
      "decode_tok_s": 75.240251528817,
      "e2e_ms": 1073.8704999675974,
      "effective_prefill_tok_s": 8288.330201345752,
      "generation_ms": 637.9563999944367,
      "mean_inter_token_latency_ms": 13.29075833321743,
      "output_token_source": "usage",
      "output_tokens": 49,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 435.9140999731608,
      "ttft_ms": 435.9143999754451,
      "visible_generation_ms": 637.9560999921523
    },
    {
      "content_chunks": 45,
      "decode_tok_s": 75.30821140384857,
      "e2e_ms": 1034.399800002575,
      "effective_prefill_tok_s": 8270.471190595268,
      "generation_ms": 597.5444000214338,
      "mean_inter_token_latency_ms": 13.278764444920752,
      "output_token_source": "usage",
      "output_tokens": 46,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 436.8553999811411,
      "ttft_ms": 436.85609998647124,
      "visible_generation_ms": 597.5437000161037
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260903T122413Z",
  "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": "2026-09-03T12:24:13Z",
  "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": 5304.81120001059
}