{
  "base_url": "http://127.0.0.1:39275/v1",
  "completed": 5,
  "concurrency": 1,
  "context_distribution": {
    "4096": 5
  },
  "context_seed": 0,
  "context_tokens": 4096,
  "engine": "sglang",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-05T03:44:07Z",
  "gpu": "Primary Node / RTX PRO 6000 Blackwell Max-Q / TP=1",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 512,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 67,
    "content_chunks_s": null,
    "decode_tok_s_max": 113.35516827666785,
    "decode_tok_s_mean": 105.09091585568599,
    "decode_tok_s_min": 95.26651741028103,
    "decode_tok_s_p25": 100.4152729431112,
    "decode_tok_s_p50": 104.95208601684021,
    "decode_tok_s_p75": 111.46553463152966,
    "decode_tok_s_p90": 113.35516827666785,
    "decode_tok_s_p95": 113.35516827666785,
    "decode_tok_s_p99": 113.35516827666785,
    "decode_tok_s_samples": 5,
    "e2e_max_ms": 600.5412000231445,
    "e2e_mean_ms": 546.4901600033045,
    "e2e_min_ms": 519.4047000259161,
    "e2e_p25_ms": 521.2005999637768,
    "e2e_p50_ms": 544.2706999601796,
    "e2e_p75_ms": 547.0336000435054,
    "e2e_p90_ms": 600.5412000231445,
    "e2e_p95_ms": 600.5412000231445,
    "e2e_p99_ms": 600.5412000231445,
    "e2e_samples": 5,
    "effective_prefill_tok_s_max": 37460.87244026109,
    "effective_prefill_tok_s_mean": 35111.62489700322,
    "effective_prefill_tok_s_min": 30700.52077870548,
    "effective_prefill_tok_s_p25": 33901.13592418131,
    "effective_prefill_tok_s_p50": 36533.84510258366,
    "effective_prefill_tok_s_p75": 36961.75023928454,
    "effective_prefill_tok_s_p90": 37460.87244026109,
    "effective_prefill_tok_s_p95": 37460.87244026109,
    "effective_prefill_tok_s_p99": 37460.87244026109,
    "effective_prefill_tok_s_samples": 5,
    "generation_max_ms": 482.85589995794,
    "generation_mean_ms": 443.01985998172313,
    "generation_min_ms": 414.6259999834001,
    "generation_p25_ms": 421.65499995462596,
    "generation_p50_ms": 447.8234000271186,
    "generation_p75_ms": 448.138999985531,
    "generation_p90_ms": 482.85589995794,
    "generation_p95_ms": 482.85589995794,
    "generation_p99_ms": 482.85589995794,
    "generation_samples": 5,
    "mean_inter_token_latency_ms_max": 10.49686739039,
    "mean_inter_token_latency_ms_mean": 9.555376409307453,
    "mean_inter_token_latency_ms_min": 8.821829786880853,
    "mean_inter_token_latency_ms_p25": 8.971382977758,
    "mean_inter_token_latency_ms_p50": 9.528157447385503,
    "mean_inter_token_latency_ms_p75": 9.95864444412291,
    "mean_inter_token_latency_ms_p90": 10.49686739039,
    "mean_inter_token_latency_ms_p95": 10.49686739039,
    "mean_inter_token_latency_ms_p99": 10.49686739039,
    "mean_inter_token_latency_ms_samples": 5,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 237,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 5,
    "prompt_tokens": 18065,
    "prompt_tokens_max": 3613.0,
    "prompt_tokens_mean": 3613.0,
    "prompt_tokens_min": 3613.0,
    "prompt_tokens_p25": 3613.0,
    "prompt_tokens_p50": 3613.0,
    "prompt_tokens_p75": 3613.0,
    "prompt_tokens_p90": 3613.0,
    "prompt_tokens_p95": 3613.0,
    "prompt_tokens_p99": 3613.0,
    "prompt_tokens_samples": 5,
    "reasoning_chunks": 0,
    "throughput_tok_s": 86.69708575989837,
    "time_to_first_output_max_ms": 117.6853000652045,
    "time_to_first_output_mean_ms": 103.47030002158135,
    "time_to_first_output_min_ms": 96.447299933061,
    "time_to_first_output_p25_ms": 97.74970007129014,
    "time_to_first_output_p50_ms": 98.89460005797446,
    "time_to_first_output_p75_ms": 106.57459998037666,
    "time_to_first_output_p90_ms": 117.6853000652045,
    "time_to_first_output_p95_ms": 117.6853000652045,
    "time_to_first_output_p99_ms": 117.6853000652045,
    "time_to_first_output_samples": 5,
    "tpot_max_ms": 10.49686739039,
    "tpot_mean_ms": 9.555376409307453,
    "tpot_min_ms": 8.821829786880853,
    "tpot_p25_ms": 8.971382977758,
    "tpot_p50_ms": 9.528157447385503,
    "tpot_p75_ms": 9.95864444412291,
    "tpot_p90_ms": 10.49686739039,
    "tpot_p95_ms": 10.49686739039,
    "tpot_p99_ms": 10.49686739039,
    "tpot_samples": 5,
    "ttft_max_ms": 117.68580006901175,
    "ttft_mean_ms": 103.47068002447486,
    "ttft_min_ms": 96.4476999361068,
    "ttft_p25_ms": 97.75010007433593,
    "ttft_p50_ms": 98.8947000587359,
    "ttft_p75_ms": 106.57509998418391,
    "ttft_p90_ms": 117.68580006901175,
    "ttft_p95_ms": 117.68580006901175,
    "ttft_p99_ms": 117.68580006901175,
    "ttft_samples": 5
  },
  "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12-chunk1k-lazy96",
  "request_timings": [
    {
      "content_chunks": 15,
      "decode_tok_s": 95.26651741028103,
      "e2e_ms": 600.5412000231445,
      "effective_prefill_tok_s": 30700.52077870548,
      "generation_ms": 482.85589995794,
      "mean_inter_token_latency_ms": 10.49686739039,
      "output_token_source": "usage",
      "output_tokens": 47,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 117.6853000652045,
      "tpot_ms": 10.49686739039,
      "ttft_ms": 117.68580006901175,
      "visible_generation_ms": 482.85539995413274
    },
    {
      "content_chunks": 13,
      "decode_tok_s": 113.35516827666785,
      "e2e_ms": 521.2005999637768,
      "effective_prefill_tok_s": 33901.13592418131,
      "generation_ms": 414.6259999834001,
      "mean_inter_token_latency_ms": 8.821829786880853,
      "output_token_source": "usage",
      "output_tokens": 48,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 106.57459998037666,
      "tpot_ms": 8.821829786880853,
      "ttft_ms": 106.57509998418391,
      "visible_generation_ms": 414.62549997959286
    },
    {
      "content_chunks": 13,
      "decode_tok_s": 100.4152729431112,
      "e2e_ms": 547.0336000435054,
      "effective_prefill_tok_s": 36533.84510258366,
      "generation_ms": 448.138999985531,
      "mean_inter_token_latency_ms": 9.95864444412291,
      "output_token_source": "usage",
      "output_tokens": 46,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 98.89460005797446,
      "tpot_ms": 9.95864444412291,
      "ttft_ms": 98.8947000587359,
      "visible_generation_ms": 448.1388999847695
    },
    {
      "content_chunks": 13,
      "decode_tok_s": 104.95208601684021,
      "e2e_ms": 544.2706999601796,
      "effective_prefill_tok_s": 37460.87244026109,
      "generation_ms": 447.8234000271186,
      "mean_inter_token_latency_ms": 9.528157447385503,
      "output_token_source": "usage",
      "output_tokens": 48,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 96.447299933061,
      "tpot_ms": 9.528157447385503,
      "ttft_ms": 96.4476999361068,
      "visible_generation_ms": 447.8230000240728
    },
    {
      "content_chunks": 13,
      "decode_tok_s": 111.46553463152966,
      "e2e_ms": 519.4047000259161,
      "effective_prefill_tok_s": 36961.75023928454,
      "generation_ms": 421.65499995462596,
      "mean_inter_token_latency_ms": 8.971382977758,
      "output_token_source": "usage",
      "output_tokens": 48,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 4,
      "time_to_first_output_ms": 97.74970007129014,
      "tpot_ms": 8.971382977758,
      "ttft_ms": 97.75010007433593,
      "visible_generation_ms": 421.65459995158017
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260905T034404Z",
  "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": "2026-09-05T03:44:04Z",
  "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",
    "distributions": "request-level mean, min, p25, p50, p75, p90, p95, p99, and max using the nearest-rank percentile method",
    "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",
    "tpot": "alias of mean_inter_token_latency for capacity-v3: client-observed generation time divided by completion tokens after the first token",
    "ttft": "request start through the first non-empty streamed content delta"
  },
  "wall_clock_ms": 2733.655899995938
}
