{
  "base_url": "http://127.0.0.1:8001/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "4096": 3
  },
  "context_seed": 0,
  "context_tokens": 4096,
  "engine": "vllm-glm53-exl3-b12x-fixed-mtp5",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-29T19:14:55Z",
  "gpu": "dual-rtx-pro-6000-tp2",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v4-reasoning",
  "metrics": {
    "content_chunks": 46,
    "content_chunks_s": null,
    "decode_tok_s_p50": 69.80057974377083,
    "decode_tok_s_p95": 76.59841747656247,
    "e2e_p50_ms": 1761.7925000013201,
    "e2e_p95_ms": 6457.352899997204,
    "effective_prefill_tok_s_p50": 3088.1040882544644,
    "effective_prefill_tok_s_p95": 3093.2555044600663,
    "generation_p50_ms": 796.361100001377,
    "generation_p95_ms": 802.2855999988678,
    "mean_inter_token_latency_ms_p50": 14.326528571408355,
    "mean_inter_token_latency_ms_p95": 15.405569565147111,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 143,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 3,
    "prompt_tokens": 8904,
    "prompt_tokens_p50": 2968.0,
    "prompt_tokens_p95": 2968.0,
    "reasoning_chunks": 6,
    "throughput_tok_s": 14.331783021935726,
    "time_to_first_output_p50_ms": 961.1074999993434,
    "time_to_first_output_p95_ms": 6103.0247999988205,
    "ttft_p50_ms": 1055.2404999980354,
    "ttft_p95_ms": 6103.025999997044,
    "ttft_samples": 3
  },
  "model": "glm53-flash-tr3-4bpw-tp2-262k-fixed-mtp5",
  "request_timings": [
    {
      "content_chunks": 9,
      "decode_tok_s": 64.91158900495029,
      "e2e_ms": 6457.352899997204,
      "effective_prefill_tok_s": 486.31622797937405,
      "generation_ms": 354.32809999838355,
      "mean_inter_token_latency_ms": 15.405569565147111,
      "output_token_source": "usage",
      "output_tokens": 24,
      "planned_context_tokens": 4096,
      "prompt_tokens": 2968,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 6103.0247999988205,
      "ttft_ms": 6103.025999997044,
      "visible_generation_ms": 354.3269000001601
    },
    {
      "content_chunks": 18,
      "decode_tok_s": 69.80057974377083,
      "e2e_ms": 1761.7925000013201,
      "effective_prefill_tok_s": 3093.2555044600663,
      "generation_ms": 802.2855999988678,
      "mean_inter_token_latency_ms": 14.326528571408355,
      "output_token_source": "usage",
      "output_tokens": 57,
      "planned_context_tokens": 4096,
      "prompt_tokens": 2968,
      "reasoning_chunks": 3,
      "request_index": 1,
      "time_to_first_output_ms": 959.5069000024523,
      "ttft_ms": 1054.327400001057,
      "visible_generation_ms": 707.4651000002632
    },
    {
      "content_chunks": 19,
      "decode_tok_s": 76.59841747656247,
      "e2e_ms": 1757.4686000007205,
      "effective_prefill_tok_s": 3088.1040882544644,
      "generation_ms": 796.361100001377,
      "mean_inter_token_latency_ms": 13.055100000022575,
      "output_token_source": "usage",
      "output_tokens": 62,
      "planned_context_tokens": 4096,
      "prompt_tokens": 2968,
      "reasoning_chunks": 3,
      "request_index": 2,
      "time_to_first_output_ms": 961.1074999993434,
      "ttft_ms": 1055.2404999980354,
      "visible_generation_ms": 702.2281000026851
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260829T191445Z",
  "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": "2026-08-29T19:14:45Z",
  "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": 9977.823400000489
}