{
  "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-k3-b12x-dflash2-k5",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-30T00:10:40Z",
  "gpu": "dual-rtx-pro-6000-max-q-tp2",
  "manifest": null,
  "max_context_tokens": 1048576,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v4-reasoning",
  "metrics": {
    "content_chunks": 79,
    "content_chunks_s": null,
    "decode_tok_s_p50": 4.256344293986114,
    "decode_tok_s_p95": 60.767909487865914,
    "e2e_p50_ms": 55213.072899998224,
    "e2e_p95_ms": 89017.2550999996,
    "effective_prefill_tok_s_p50": 1800.743889949181,
    "effective_prefill_tok_s_p95": 2431.2984017961,
    "generation_p50_ms": 11332.532199994603,
    "generation_p95_ms": 15192.113600001903,
    "mean_inter_token_latency_ms_p50": 222.20651372538435,
    "mean_inter_token_latency_ms_p95": 330.26333913047614,
    "output_token_sources": {
      "usage": 4
    },
    "output_tokens": 200,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 4,
    "prompt_tokens": 422661,
    "prompt_tokens_p50": 105664.0,
    "prompt_tokens_p95": 105668.0,
    "reasoning_chunks": 5,
    "throughput_tok_s": 1.3866226304160962,
    "time_to_first_output_p50_ms": 54390.27019999776,
    "time_to_first_output_p95_ms": 73825.1414999977,
    "ttft_p50_ms": 54390.27080000233,
    "ttft_p95_ms": 82274.45360000274,
    "ttft_samples": 4
  },
  "model": "glm53-flash-exl3-k3-dflash2-k5-fp8-tp2-1m-vision",
  "request_timings": [
    {
      "content_chunks": 13,
      "decode_tok_s": 3.0278867846271393,
      "e2e_ms": 89017.2550999996,
      "effective_prefill_tok_s": 1431.2603789591558,
      "generation_ms": 15192.113600001903,
      "mean_inter_token_latency_ms": 330.26333913047614,
      "output_token_source": "usage",
      "output_tokens": 47,
      "planned_context_tokens": 131072,
      "prompt_tokens": 105663,
      "reasoning_chunks": 5,
      "request_index": 0,
      "time_to_first_output_ms": 73825.1414999977,
      "ttft_ms": 82274.45360000274,
      "visible_generation_ms": 6742.801499996858
    },
    {
      "content_chunks": 21,
      "decode_tok_s": 4.256344293986114,
      "e2e_ms": 54972.132599999895,
      "effective_prefill_tok_s": 2431.2984017961,
      "generation_ms": 11512.226600003487,
      "mean_inter_token_latency_ms": 234.94340000007116,
      "output_token_source": "usage",
      "output_tokens": 50,
      "planned_context_tokens": 131072,
      "prompt_tokens": 105664,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 43459.90599999641,
      "ttft_ms": 43459.90699999675,
      "visible_generation_ms": 11512.225600003148
    },
    {
      "content_chunks": 21,
      "decode_tok_s": 4.500318119548276,
      "e2e_ms": 70011.61509999656,
      "effective_prefill_tok_s": 1800.743889949181,
      "generation_ms": 11332.532199994603,
      "mean_inter_token_latency_ms": 222.20651372538435,
      "output_token_source": "usage",
      "output_tokens": 52,
      "planned_context_tokens": 131072,
      "prompt_tokens": 105666,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 58679.08290000196,
      "ttft_ms": 58679.08429999807,
      "visible_generation_ms": 11332.530799998494
    },
    {
      "content_chunks": 24,
      "decode_tok_s": 60.767909487865914,
      "e2e_ms": 55213.072899998224,
      "effective_prefill_tok_s": 1942.7739485656086,
      "generation_ms": 822.8027000004658,
      "mean_inter_token_latency_ms": 16.456054000009317,
      "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": 54390.27019999776,
      "ttft_ms": 54390.27080000233,
      "visible_generation_ms": 822.8020999958972
    }
  ],
  "requests": 4,
  "run_id": "benchmark-20260830T000816Z",
  "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": "2026-08-30T00:08:16Z",
  "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": 144235.34970000037
}