{
  "base_url": "http://127.0.0.1:8001/v1",
  "completed": 4,
  "concurrency": 2,
  "context_distribution": {
    "131072": 4
  },
  "context_seed": 0,
  "context_tokens": 131072,
  "engine": "vllm-glm53-exl3-b12x-fixed-mtp5-vision",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-29T22:32:36Z",
  "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": 74,
    "content_chunks_s": null,
    "decode_tok_s_p50": 4.567389880230011,
    "decode_tok_s_p95": 52.172749817984474,
    "e2e_p50_ms": 56559.40589999955,
    "e2e_p95_ms": 80011.23609999922,
    "effective_prefill_tok_s_p50": 1892.2712360235853,
    "effective_prefill_tok_s_p95": 2542.878174071721,
    "generation_p50_ms": 8587.51040000061,
    "generation_p95_ms": 10728.227999999945,
    "mean_inter_token_latency_ms_p50": 185.56163584901884,
    "mean_inter_token_latency_ms_p95": 220.19257435899002,
    "output_token_sources": {
      "usage": 4
    },
    "output_tokens": 195,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 4,
    "prompt_tokens": 422661,
    "prompt_tokens_p50": 105664.0,
    "prompt_tokens_p95": 105668.0,
    "reasoning_chunks": 4,
    "throughput_tok_s": 1.4277624855426192,
    "time_to_first_output_p50_ms": 55601.05119999935,
    "time_to_first_output_p95_ms": 71423.72569999861,
    "ttft_p50_ms": 55601.05160000239,
    "ttft_p95_ms": 71423.72839999734,
    "ttft_samples": 4
  },
  "model": "glm53-flash-tr3-4bpw-tp2-262k-fixed-mtp5-vision",
  "request_timings": [
    {
      "content_chunks": 15,
      "decode_tok_s": 5.389044968398771,
      "e2e_ms": 51387.2881999996,
      "effective_prefill_tok_s": 2542.878174071721,
      "generation_ms": 9834.766699998,
      "mean_inter_token_latency_ms": 185.56163584901884,
      "output_token_source": "usage",
      "output_tokens": 54,
      "planned_context_tokens": 131072,
      "prompt_tokens": 105663,
      "reasoning_chunks": 4,
      "request_index": 0,
      "time_to_first_output_ms": 41552.5215000016,
      "ttft_ms": 43347.010900004534,
      "visible_generation_ms": 8040.2772999950685
    },
    {
      "content_chunks": 17,
      "decode_tok_s": 4.541479216141296,
      "e2e_ms": 80011.23609999922,
      "effective_prefill_tok_s": 1479.3963625451431,
      "generation_ms": 8587.51040000061,
      "mean_inter_token_latency_ms": 220.19257435899002,
      "output_token_source": "usage",
      "output_tokens": 40,
      "planned_context_tokens": 131072,
      "prompt_tokens": 105664,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 71423.72569999861,
      "ttft_ms": 71423.72839999734,
      "visible_generation_ms": 8587.50770000188
    },
    {
      "content_chunks": 21,
      "decode_tok_s": 4.567389880230011,
      "e2e_ms": 66569.05990000087,
      "effective_prefill_tok_s": 1892.2712360235853,
      "generation_ms": 10728.227999999945,
      "mean_inter_token_latency_ms": 218.94342857142743,
      "output_token_source": "usage",
      "output_tokens": 50,
      "planned_context_tokens": 131072,
      "prompt_tokens": 105666,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 55840.83190000092,
      "ttft_ms": 55840.8328000005,
      "visible_generation_ms": 10728.227100000367
    },
    {
      "content_chunks": 21,
      "decode_tok_s": 52.172749817984474,
      "e2e_ms": 56559.40589999955,
      "effective_prefill_tok_s": 1900.4676659782513,
      "generation_ms": 958.3547000002,
      "mean_inter_token_latency_ms": 19.167094000004,
      "output_token_source": "usage",
      "output_tokens": 51,
      "planned_context_tokens": 131072,
      "prompt_tokens": 105668,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 55601.05119999935,
      "ttft_ms": 55601.05160000239,
      "visible_generation_ms": 958.3542999971542
    }
  ],
  "requests": 4,
  "run_id": "benchmark-20260829T223019Z",
  "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": "2026-08-29T22:30:19Z",
  "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": 136577.3382999978
}