{
  "base_url": "http://127.0.0.1:39038/v1",
  "completed": 10,
  "concurrency": 1,
  "context_distribution": {
    "4096": 10
  },
  "context_seed": 0,
  "context_tokens": 4096,
  "engine": "sglang",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-21T14:44:43Z",
  "gpu": "rtx-5090",
  "manifest": null,
  "max_context_tokens": 131072,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 460,
    "content_chunks_s": null,
    "decode_tok_s_p50": 76.53764121009965,
    "decode_tok_s_p95": 76.77077672442144,
    "e2e_p50_ms": 911.0958999954164,
    "e2e_p95_ms": 1252.9303999617696,
    "effective_prefill_tok_s_p50": 11159.990339843103,
    "effective_prefill_tok_s_p95": 11259.720585623007,
    "generation_p50_ms": 588.2553999545053,
    "generation_p95_ms": 640.4937000479549,
    "mean_inter_token_latency_ms_p50": 13.064325581369705,
    "mean_inter_token_latency_ms_p95": 13.072872915169379,
    "output_token_sources": {
      "usage": 10
    },
    "output_tokens": 470,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 10,
    "prompt_tokens": 36130,
    "prompt_tokens_p50": 3613.0,
    "prompt_tokens_p95": 3613.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 49.30263158410785,
    "time_to_first_output_p50_ms": 323.4427000861615,
    "time_to_first_output_p95_ms": 614.6667000139132,
    "ttft_p50_ms": 323.4431000892073,
    "ttft_p95_ms": 614.6675999043509,
    "ttft_samples": 10
  },
  "model": "qwen38-27b-radixark-nvfp4-sglang-rtx5090-128k-mm",
  "request_timings": [
    {
      "content_chunks": 49,
      "decode_tok_s": 76.77077672442144,
      "e2e_ms": 1252.9303999617696,
      "effective_prefill_tok_s": 5877.982327525175,
      "generation_ms": 638.2636999478564,
      "mean_inter_token_latency_ms": 13.025789794854212,
      "output_token_source": "usage",
      "output_tokens": 50,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 614.6667000139132,
      "ttft_ms": 614.6675999043509,
      "visible_generation_ms": 638.2628000574186
    },
    {
      "content_chunks": 43,
      "decode_tok_s": 76.5443262854719,
      "e2e_ms": 882.6442999998108,
      "effective_prefill_tok_s": 11259.720585623007,
      "generation_ms": 561.7659999988973,
      "mean_inter_token_latency_ms": 13.064325581369705,
      "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": 320.87830000091344,
      "ttft_ms": 320.87890000548214,
      "visible_generation_ms": 561.7653999943286
    },
    {
      "content_chunks": 48,
      "decode_tok_s": 76.56109263359829,
      "e2e_ms": 947.9741000104696,
      "effective_prefill_tok_s": 11254.617260847106,
      "generation_ms": 626.9503000658005,
      "mean_inter_token_latency_ms": 13.061464584704177,
      "output_token_source": "usage",
      "output_tokens": 49,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 321.023799944669,
      "ttft_ms": 321.02429994847625,
      "visible_generation_ms": 626.9498000619933
    },
    {
      "content_chunks": 43,
      "decode_tok_s": 76.56151210463211,
      "e2e_ms": 883.9626999106258,
      "effective_prefill_tok_s": 11209.25978759772,
      "generation_ms": 561.6398999700323,
      "mean_inter_token_latency_ms": 13.06139302255889,
      "output_token_source": "usage",
      "output_tokens": 44,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 322.32279994059354,
      "ttft_ms": 322.3232999444008,
      "visible_generation_ms": 561.639399966225
    },
    {
      "content_chunks": 45,
      "decode_tok_s": 76.49738532528598,
      "e2e_ms": 911.0958999954164,
      "effective_prefill_tok_s": 11191.284859062453,
      "generation_ms": 588.2553999545053,
      "mean_inter_token_latency_ms": 13.072342221211228,
      "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": 322.84050004091114,
      "ttft_ms": 322.8410000447184,
      "visible_generation_ms": 588.254899950698
    },
    {
      "content_chunks": 43,
      "decode_tok_s": 76.50296751682566,
      "e2e_ms": 885.8918000478297,
      "effective_prefill_tok_s": 11157.360783213766,
      "generation_ms": 562.0696999831125,
      "mean_inter_token_latency_ms": 13.07138837170029,
      "output_token_source": "usage",
      "output_tokens": 44,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 5,
      "time_to_first_output_ms": 323.82210006471723,
      "ttft_ms": 323.8224000670016,
      "visible_generation_ms": 562.0693999808282
    },
    {
      "content_chunks": 48,
      "decode_tok_s": 76.49427990993695,
      "e2e_ms": 950.9406000142917,
      "effective_prefill_tok_s": 11170.448425756826,
      "generation_ms": 627.4978999281302,
      "mean_inter_token_latency_ms": 13.072872915169379,
      "output_token_source": "usage",
      "output_tokens": 49,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 6,
      "time_to_first_output_ms": 323.4427000861615,
      "ttft_ms": 323.4431000892073,
      "visible_generation_ms": 627.4974999250844
    },
    {
      "content_chunks": 44,
      "decode_tok_s": 76.54781864273745,
      "e2e_ms": 899.1078999824822,
      "effective_prefill_tok_s": 11140.788359304936,
      "generation_ms": 574.8040999751538,
      "mean_inter_token_latency_ms": 13.06372954488986,
      "output_token_source": "usage",
      "output_tokens": 45,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 7,
      "time_to_first_output_ms": 324.3038000073284,
      "ttft_ms": 324.30450001265854,
      "visible_generation_ms": 574.8033999698237
    },
    {
      "content_chunks": 48,
      "decode_tok_s": 76.53764121009965,
      "e2e_ms": 950.8881999645382,
      "effective_prefill_tok_s": 11159.990339843103,
      "generation_ms": 627.1424000151455,
      "mean_inter_token_latency_ms": 13.065466666982198,
      "output_token_source": "usage",
      "output_tokens": 49,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 8,
      "time_to_first_output_ms": 323.7457999493927,
      "ttft_ms": 323.7462999531999,
      "visible_generation_ms": 627.1419000113383
    },
    {
      "content_chunks": 49,
      "decode_tok_s": 76.50348472800167,
      "e2e_ms": 965.3704001102597,
      "effective_prefill_tok_s": 11121.142265071949,
      "generation_ms": 640.4937000479549,
      "mean_inter_token_latency_ms": 13.071300000978672,
      "output_token_source": "usage",
      "output_tokens": 50,
      "planned_context_tokens": 4096,
      "prompt_tokens": 3613,
      "reasoning_chunks": 0,
      "request_index": 9,
      "time_to_first_output_ms": 324.87670006230474,
      "ttft_ms": 324.877200066112,
      "visible_generation_ms": 640.4932000441477
    }
  ],
  "requests": 10,
  "run_id": "benchmark-20260821T144434Z",
  "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": "2026-08-21T14:44:34Z",
  "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": 9532.959700096399
}