{
  "base_url": "http://127.0.0.1:8001/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "128000": 3
  },
  "context_seed": 0,
  "context_tokens": 128000,
  "engine": "vllm-glm53-exl3-b12x-fixed-mtp5",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-29T19:16:47Z",
  "gpu": "dual-rtx-pro-6000-tp2",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 53,
    "content_chunks_s": null,
    "decode_tok_s_p50": 61.85295809853271,
    "decode_tok_s_p95": 72.34537524286536,
    "e2e_p50_ms": 33690.131100000144,
    "e2e_p95_ms": 38856.07700000037,
    "effective_prefill_tok_s_p50": 3132.902311479185,
    "effective_prefill_tok_s_p95": 3147.4977669124773,
    "generation_p50_ms": 760.2421000010509,
    "generation_p95_ms": 792.2014000032505,
    "mean_inter_token_latency_ms_p50": 16.167375510270418,
    "mean_inter_token_latency_ms_p95": 17.730866666641777,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 134,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 3,
    "prompt_tokens": 309199,
    "prompt_tokens_p50": 103066.0,
    "prompt_tokens_p95": 103067.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 1.2634581725930134,
    "time_to_first_output_p50_ms": 32897.929699996894,
    "time_to_first_output_p95_ms": 38377.34360000104,
    "ttft_p50_ms": 32897.930399998586,
    "ttft_p95_ms": 38377.344399999856,
    "ttft_samples": 3
  },
  "model": "glm53-flash-tr3-4bpw-tp2-262k-fixed-mtp5",
  "request_timings": [
    {
      "content_chunks": 12,
      "decode_tok_s": 56.39882239266762,
      "e2e_ms": 38856.07700000037,
      "effective_prefill_tok_s": 2685.5949456594803,
      "generation_ms": 478.73339999932796,
      "mean_inter_token_latency_ms": 17.730866666641777,
      "output_token_source": "usage",
      "output_tokens": 28,
      "planned_context_tokens": 128000,
      "prompt_tokens": 103066,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 38377.34360000104,
      "ttft_ms": 38377.344399999856,
      "visible_generation_ms": 478.7326000005123
    },
    {
      "content_chunks": 20,
      "decode_tok_s": 72.34537524286536,
      "e2e_ms": 33505.93650000155,
      "effective_prefill_tok_s": 3147.4977669124773,
      "generation_ms": 760.2421000010509,
      "mean_inter_token_latency_ms": 13.822583636382744,
      "output_token_source": "usage",
      "output_tokens": 56,
      "planned_context_tokens": 128000,
      "prompt_tokens": 103067,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 32745.694400000502,
      "ttft_ms": 32745.695100002195,
      "visible_generation_ms": 760.2413999993587
    },
    {
      "content_chunks": 21,
      "decode_tok_s": 61.85295809853271,
      "e2e_ms": 33690.131100000144,
      "effective_prefill_tok_s": 3132.902311479185,
      "generation_ms": 792.2014000032505,
      "mean_inter_token_latency_ms": 16.167375510270418,
      "output_token_source": "usage",
      "output_tokens": 50,
      "planned_context_tokens": 128000,
      "prompt_tokens": 103066,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 32897.929699996894,
      "ttft_ms": 32897.930399998586,
      "visible_generation_ms": 792.2007000015583
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260829T191501Z",
  "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": "2026-08-29T19:15:01Z",
  "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": 106058.12119999973
}