{
  "base_url": "http://127.0.0.1:8001/v1",
  "completed": 5,
  "concurrency": 1,
  "context_distribution": {
    "65536": 5
  },
  "context_seed": 0,
  "context_tokens": 65536,
  "engine": "vllm-glm53-exl3-b12x-fixed-mtp5-vision",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-29T22:20:23Z",
  "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": 94,
    "content_chunks_s": null,
    "decode_tok_s_p50": 55.24569972508946,
    "decode_tok_s_p95": 66.01962798291218,
    "e2e_p50_ms": 17435.162499998114,
    "e2e_p95_ms": 17622.016199995414,
    "effective_prefill_tok_s_p50": 3075.0121388220073,
    "effective_prefill_tok_s_p95": 3104.714584061704,
    "generation_p50_ms": 995.5525999976089,
    "generation_p95_ms": 1226.401599997189,
    "mean_inter_token_latency_ms_p50": 18.10095636359289,
    "mean_inter_token_latency_ms_p95": 20.440026666619815,
    "output_token_sources": {
      "usage": 5
    },
    "output_tokens": 293,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 5,
    "prompt_tokens": 252752,
    "prompt_tokens_p50": 50551.0,
    "prompt_tokens_p95": 50553.0,
    "reasoning_chunks": 22,
    "throughput_tok_s": 3.3600938046448174,
    "time_to_first_output_p50_ms": 16439.609900000505,
    "time_to_first_output_p95_ms": 16483.290899996064,
    "ttft_p50_ms": 16596.995799998695,
    "ttft_p95_ms": 16682.359599995834,
    "ttft_samples": 5
  },
  "model": "glm53-flash-tr3-4bpw-tp2-262k-fixed-mtp5-vision",
  "request_timings": [
    {
      "content_chunks": 23,
      "decode_tok_s": 48.92361523349083,
      "e2e_ms": 17507.125199997972,
      "effective_prefill_tok_s": 3104.714584061704,
      "generation_ms": 1226.401599997189,
      "mean_inter_token_latency_ms": 20.440026666619815,
      "output_token_source": "usage",
      "output_tokens": 61,
      "planned_context_tokens": 65536,
      "prompt_tokens": 50547,
      "reasoning_chunks": 4,
      "request_index": 0,
      "time_to_first_output_ms": 16280.723600000783,
      "ttft_ms": 16426.827099996444,
      "visible_generation_ms": 1080.298100001528
    },
    {
      "content_chunks": 18,
      "decode_tok_s": 51.77996744905216,
      "e2e_ms": 17321.63810000202,
      "effective_prefill_tok_s": 3090.5452225191725,
      "generation_ms": 965.6243999997969,
      "mean_inter_token_latency_ms": 19.312487999995938,
      "output_token_source": "usage",
      "output_tokens": 51,
      "planned_context_tokens": 65536,
      "prompt_tokens": 50549,
      "reasoning_chunks": 4,
      "request_index": 1,
      "time_to_first_output_ms": 16356.013700002222,
      "ttft_ms": 16512.25919999706,
      "visible_generation_ms": 809.3789000049583
    },
    {
      "content_chunks": 21,
      "decode_tok_s": 57.95954476469227,
      "e2e_ms": 17622.016199995414,
      "effective_prefill_tok_s": 3066.8026370882085,
      "generation_ms": 1138.7252999993507,
      "mean_inter_token_latency_ms": 17.253413636353798,
      "output_token_source": "usage",
      "output_tokens": 67,
      "planned_context_tokens": 65536,
      "prompt_tokens": 50551,
      "reasoning_chunks": 5,
      "request_index": 2,
      "time_to_first_output_ms": 16483.290899996064,
      "ttft_ms": 16682.359599995834,
      "visible_generation_ms": 939.6565999995801
    },
    {
      "content_chunks": 17,
      "decode_tok_s": 55.24569972508946,
      "e2e_ms": 17435.162499998114,
      "effective_prefill_tok_s": 3075.0121388220073,
      "generation_ms": 995.5525999976089,
      "mean_inter_token_latency_ms": 18.10095636359289,
      "output_token_source": "usage",
      "output_tokens": 56,
      "planned_context_tokens": 65536,
      "prompt_tokens": 50552,
      "reasoning_chunks": 5,
      "request_index": 3,
      "time_to_first_output_ms": 16439.609900000505,
      "ttft_ms": 16635.084699999425,
      "visible_generation_ms": 800.0777999986894
    },
    {
      "content_chunks": 15,
      "decode_tok_s": 66.01962798291218,
      "e2e_ms": 17308.491000003414,
      "effective_prefill_tok_s": 3074.044241499504,
      "generation_ms": 863.3795999994618,
      "mean_inter_token_latency_ms": 15.147010526306348,
      "output_token_source": "usage",
      "output_tokens": 58,
      "planned_context_tokens": 65536,
      "prompt_tokens": 50553,
      "reasoning_chunks": 4,
      "request_index": 4,
      "time_to_first_output_ms": 16445.111400003952,
      "ttft_ms": 16596.995799998695,
      "visible_generation_ms": 711.4952000047197
    }
  ],
  "requests": 5,
  "run_id": "benchmark-20260829T221856Z",
  "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": "2026-08-29T22:18:56Z",
  "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": 87199.94649999717
}