{
  "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-b12x-fixed-mtp5-vision",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-29T22:18:56Z",
  "gpu": "dual-rtx-pro-6000-max-q-tp2",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v4-reasoning",
  "metrics": {
    "content_chunks": 91,
    "content_chunks_s": null,
    "decode_tok_s_p50": 50.50942831787052,
    "decode_tok_s_p95": 56.69901723584371,
    "e2e_p50_ms": 8947.783000003255,
    "e2e_p95_ms": 9343.34939999826,
    "effective_prefill_tok_s_p50": 3089.074791916423,
    "effective_prefill_tok_s_p95": 3116.064951189037,
    "generation_p50_ms": 881.8495000014082,
    "generation_p95_ms": 1242.8656999982195,
    "mean_inter_token_latency_ms_p50": 19.79828387101729,
    "mean_inter_token_latency_ms_p95": 21.09870000008653,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 243,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 5,
    "prompt_tokens": 125115,
    "prompt_tokens_p50": 25023.0,
    "prompt_tokens_p95": 25024.0,
    "reasoning_chunks": 11,
    "throughput_tok_s": 5.392730658991027,
    "time_to_first_output_p50_ms": 8100.483700000041,
    "time_to_first_output_p95_ms": 8146.032399999967,
    "ttft_p50_ms": 8186.557799999719,
    "ttft_p95_ms": 8255.982700000459,
    "ttft_samples": 5
  },
  "model": "glm53-flash-tr3-4bpw-tp2-262k-fixed-mtp5-vision",
  "request_timings": [
    {
      "content_chunks": 16,
      "decode_tok_s": 56.69901723584371,
      "e2e_ms": 8911.848999996437,
      "effective_prefill_tok_s": 3116.064951189037,
      "generation_ms": 881.8495000014082,
      "mean_inter_token_latency_ms": 17.636990000028163,
      "output_token_source": "usage",
      "output_tokens": 51,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25022,
      "reasoning_chunks": 4,
      "request_index": 0,
      "time_to_first_output_ms": 8029.999499995029,
      "ttft_ms": 8186.557799999719,
      "visible_generation_ms": 725.2911999967182
    },
    {
      "content_chunks": 14,
      "decode_tok_s": 50.50942831787052,
      "e2e_ms": 8753.217800003767,
      "effective_prefill_tok_s": 3074.2784144071697,
      "generation_ms": 613.746800001536,
      "mean_inter_token_latency_ms": 19.79828387101729,
      "output_token_source": "usage",
      "output_tokens": 32,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25023,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 8139.471000002231,
      "ttft_ms": 8139.473200004431,
      "visible_generation_ms": 613.7445999993361
    },
    {
      "content_chunks": 17,
      "decode_tok_s": 47.396285078981116,
      "e2e_ms": 8947.783000003255,
      "effective_prefill_tok_s": 3071.8021696059177,
      "generation_ms": 801.7506000032881,
      "mean_inter_token_latency_ms": 21.09870000008653,
      "output_token_source": "usage",
      "output_tokens": 39,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25023,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 8146.032399999967,
      "ttft_ms": 8146.03330000682,
      "visible_generation_ms": 801.7496999964351
    },
    {
      "content_chunks": 20,
      "decode_tok_s": 49.4750597213904,
      "e2e_ms": 9101.343700000143,
      "effective_prefill_tok_s": 3092.921055921155,
      "generation_ms": 1010.6101999990642,
      "mean_inter_token_latency_ms": 20.212203999981284,
      "output_token_source": "usage",
      "output_tokens": 51,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25024,
      "reasoning_chunks": 3,
      "request_index": 3,
      "time_to_first_output_ms": 8090.733500001079,
      "ttft_ms": 8194.186700005957,
      "visible_generation_ms": 907.1569999941858
    },
    {
      "content_chunks": 24,
      "decode_tok_s": 55.51685914262406,
      "e2e_ms": 9343.34939999826,
      "effective_prefill_tok_s": 3089.074791916423,
      "generation_ms": 1242.8656999982195,
      "mean_inter_token_latency_ms": 18.01254637678579,
      "output_token_source": "usage",
      "output_tokens": 70,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25023,
      "reasoning_chunks": 4,
      "request_index": 4,
      "time_to_first_output_ms": 8100.483700000041,
      "ttft_ms": 8255.982700000459,
      "visible_generation_ms": 1087.3666999978013
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260829T221811Z",
  "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": "2026-08-29T22:18:11Z",
  "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": 45060.65949999902
}