{
  "base_url": "http://127.0.0.1:8001/v1",
  "completed": 5,
  "concurrency": 1,
  "context_distribution": {
    "32768": 5
  },
  "context_seed": 0,
  "context_tokens": 32768,
  "engine": "vllm-glm53-exl3-k3-b12x-dflash2-k5",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-29T23:27:09Z",
  "gpu": "dual-rtx-pro-6000-max-q-tp2",
  "manifest": null,
  "max_context_tokens": 1048576,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v4-reasoning",
  "metrics": {
    "content_chunks": 99,
    "content_chunks_s": null,
    "decode_tok_s_p50": 62.24030418306764,
    "decode_tok_s_p95": 75.96764773627747,
    "e2e_p50_ms": 8499.515800001973,
    "e2e_p95_ms": 8622.701000000234,
    "effective_prefill_tok_s_p50": 3221.6171333587654,
    "effective_prefill_tok_s_p95": 3252.5340597759196,
    "generation_p50_ms": 763.4829000016907,
    "generation_p95_ms": 868.0256000006921,
    "mean_inter_token_latency_ms_p50": 16.066759523839988,
    "mean_inter_token_latency_ms_p95": 16.182909999915864,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 249,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 5,
    "prompt_tokens": 125115,
    "prompt_tokens_p50": 25023.0,
    "prompt_tokens_p95": 25024.0,
    "reasoning_chunks": 8,
    "throughput_tok_s": 5.846598745406443,
    "time_to_first_output_p50_ms": 7767.5276000009035,
    "time_to_first_output_p95_ms": 7894.503399998939,
    "ttft_p50_ms": 7813.558300003933,
    "ttft_p95_ms": 7958.9440999989165,
    "ttft_samples": 5
  },
  "model": "glm53-flash-exl3-k3-dflash2-k5-fp8-tp2-1m-vision",
  "request_timings": [
    {
      "content_chunks": 14,
      "decode_tok_s": 70.8554645949181,
      "e2e_ms": 8459.03289999842,
      "effective_prefill_tok_s": 3169.547054727136,
      "generation_ms": 564.5294999994803,
      "mean_inter_token_latency_ms": 14.113237499987008,
      "output_token_source": "usage",
      "output_tokens": 41,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25022,
      "reasoning_chunks": 2,
      "request_index": 0,
      "time_to_first_output_ms": 7894.503399998939,
      "ttft_ms": 7958.9440999989165,
      "visible_generation_ms": 500.08879999950295
    },
    {
      "content_chunks": 23,
      "decode_tok_s": 62.210146797464205,
      "e2e_ms": 8561.411599999701,
      "effective_prefill_tok_s": 3252.5340597759196,
      "generation_ms": 868.0256000006921,
      "mean_inter_token_latency_ms": 16.074548148160964,
      "output_token_source": "usage",
      "output_tokens": 55,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25023,
      "reasoning_chunks": 3,
      "request_index": 1,
      "time_to_first_output_ms": 7693.385999999009,
      "ttft_ms": 7766.682200002833,
      "visible_generation_ms": 794.7293999968679
    },
    {
      "content_chunks": 24,
      "decode_tok_s": 61.79358347820009,
      "e2e_ms": 8622.701000000234,
      "effective_prefill_tok_s": 3202.511327907734,
      "generation_ms": 809.1454999957932,
      "mean_inter_token_latency_ms": 16.182909999915864,
      "output_token_source": "usage",
      "output_tokens": 51,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25023,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 7813.555500004441,
      "ttft_ms": 7813.558300003933,
      "visible_generation_ms": 809.1426999963005
    },
    {
      "content_chunks": 19,
      "decode_tok_s": 62.24030418306764,
      "e2e_ms": 8442.331500002183,
      "effective_prefill_tok_s": 3221.6171333587654,
      "generation_ms": 674.8039000012795,
      "mean_inter_token_latency_ms": 16.066759523839988,
      "output_token_source": "usage",
      "output_tokens": 43,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25024,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 7767.5276000009035,
      "ttft_ms": 7767.528299998958,
      "visible_generation_ms": 674.8032000032254
    },
    {
      "content_chunks": 19,
      "decode_tok_s": 75.96764773627747,
      "e2e_ms": 8499.515800001973,
      "effective_prefill_tok_s": 3234.603617055337,
      "generation_ms": 763.4829000016907,
      "mean_inter_token_latency_ms": 13.16349827589122,
      "output_token_source": "usage",
      "output_tokens": 59,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25023,
      "reasoning_chunks": 3,
      "request_index": 4,
      "time_to_first_output_ms": 7736.032900000282,
      "ttft_ms": 7835.999700000684,
      "visible_generation_ms": 663.5161000012886
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260829T232626Z",
  "schema": "anvil-serving.benchmark/v1",
  "serve_flags": {
    "no_thinking": false,
    "reasoning_effort": "low",
    "shared_prefix_burst": false,
    "thinking_mode": "default"
  },
  "source_recipe": null,
  "started_at": "2026-08-29T23:26:26Z",
  "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": 42588.86419999908
}