{
  "base_url": "http://127.0.0.1:8081/v1",
  "completed": 5,
  "concurrency": 1,
  "context_distribution": {
    "4096": 5
  },
  "context_seed": 0,
  "context_tokens": 4096,
  "engine": "sglang",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-03T18:54:37Z",
  "gpu": "rtx-5090",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 216,
    "content_chunks_s": null,
    "decode_tok_s_p50": 79.48655658309721,
    "decode_tok_s_p95": 79.56017544193318,
    "e2e_p50_ms": 590.4653000179678,
    "e2e_p95_ms": 830.0583999953233,
    "effective_prefill_tok_s_p50": 70920.31529722911,
    "effective_prefill_tok_s_p95": 73370.23837668217,
    "generation_p50_ms": 541.0528000211343,
    "generation_p95_ms": 603.8757000351325,
    "mean_inter_token_latency_ms_p50": 12.580743750731926,
    "mean_inter_token_latency_ms_p95": 12.589212194675715,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 221,
    "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": 68.96205891638543,
    "time_to_first_output_p50_ms": 50.9444999624975,
    "time_to_first_output_p95_ms": 276.9526999909431,
    "ttft_p50_ms": 50.94519996782765,
    "ttft_p95_ms": 276.95329999551177,
    "ttft_samples": 5
  },
  "model": "qwen38-27b-gittensor-nvfp4-nospec-262k",
  "request_timings": [
    {
      "content_chunks": 44,
      "decode_tok_s": 79.55079833683065,
      "e2e_ms": 830.0583999953233,
      "effective_prefill_tok_s": 13045.54893351158,
      "generation_ms": 553.1057000043802,
      "mean_inter_token_latency_ms": 12.57058409100864,
      "output_token_source": "usage",
      "output_tokens": 45,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 276.9526999909431,
      "ttft_ms": 276.95329999551177,
      "visible_generation_ms": 553.1050999998115
    },
    {
      "content_chunks": 48,
      "decode_tok_s": 79.48655658309721,
      "e2e_ms": 654.82019999763,
      "effective_prefill_tok_s": 70920.31529722911,
      "generation_ms": 603.8757000351325,
      "mean_inter_token_latency_ms": 12.580743750731926,
      "output_token_source": "usage",
      "output_tokens": 49,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 50.9444999624975,
      "ttft_ms": 50.94519996782765,
      "visible_generation_ms": 603.8750000298023
    },
    {
      "content_chunks": 41,
      "decode_tok_s": 79.43308799123461,
      "e2e_ms": 575.7204000256024,
      "effective_prefill_tok_s": 60658.767942641905,
      "generation_ms": 516.1576999817044,
      "mean_inter_token_latency_ms": 12.589212194675715,
      "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": 59.562700043898076,
      "ttft_ms": 59.56339999102056,
      "visible_generation_ms": 516.1570000345819
    },
    {
      "content_chunks": 40,
      "decode_tok_s": 79.56017544193318,
      "e2e_ms": 552.0075000240467,
      "effective_prefill_tok_s": 73370.23837668217,
      "generation_ms": 502.7640999760479,
      "mean_inter_token_latency_ms": 12.569102499401197,
      "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": 49.243400047998875,
      "ttft_ms": 49.24380005104467,
      "visible_generation_ms": 502.7636999730021
    },
    {
      "content_chunks": 43,
      "decode_tok_s": 79.47468342890076,
      "e2e_ms": 590.4653000179678,
      "effective_prefill_tok_s": 73119.15001733431,
      "generation_ms": 541.0528000211343,
      "mean_inter_token_latency_ms": 12.58262325630545,
      "output_token_source": "usage",
      "output_tokens": 44,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 49.4124999968335,
      "ttft_ms": 49.41300000064075,
      "visible_generation_ms": 541.0523000173271
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260903T185434Z",
  "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": "2026-09-03T18:54:34Z",
  "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": 3204.6606999938376
}