{
  "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-vision",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-29T20:38:38Z",
  "gpu": "dual-rtx-pro-6000-tp2",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v4-reasoning",
  "metrics": {
    "content_chunks": 37,
    "content_chunks_s": null,
    "decode_tok_s_p50": 72.8246995804758,
    "decode_tok_s_p95": 74.46196267841451,
    "e2e_p50_ms": 1571.340099995723,
    "e2e_p95_ms": 6451.7102999998315,
    "effective_prefill_tok_s_p50": 2943.439061449323,
    "effective_prefill_tok_s_p95": 3100.5031005130127,
    "generation_p50_ms": 562.9957999990438,
    "generation_p95_ms": 604.3353999993997,
    "mean_inter_token_latency_ms_p50": 13.73160487802546,
    "mean_inter_token_latency_ms_p95": 13.888356521657323,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 112,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 3,
    "prompt_tokens": 8904,
    "prompt_tokens_p50": 2968.0,
    "prompt_tokens_p95": 2968.0,
    "reasoning_chunks": 3,
    "throughput_tok_s": 11.683599732102406,
    "time_to_first_output_p50_ms": 1008.3442999966792,
    "time_to_first_output_p95_ms": 6132.278100001713,
    "ttft_p50_ms": 1048.4871000007843,
    "ttft_p95_ms": 6132.279200002813,
    "ttft_samples": 3
  },
  "model": "glm53-flash-tr3-4bpw-tp2-262k-fixed-mtp5-vision",
  "request_timings": [
    {
      "content_chunks": 8,
      "decode_tok_s": 72.00275989751653,
      "e2e_ms": 6451.7102999998315,
      "effective_prefill_tok_s": 483.99631451795557,
      "generation_ms": 319.43219999811845,
      "mean_inter_token_latency_ms": 13.888356521657323,
      "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": 6132.278100001713,
      "ttft_ms": 6132.279200002813,
      "visible_generation_ms": 319.43109999701846
    },
    {
      "content_chunks": 15,
      "decode_tok_s": 72.8246995804758,
      "e2e_ms": 1571.340099995723,
      "effective_prefill_tok_s": 2943.439061449323,
      "generation_ms": 562.9957999990438,
      "mean_inter_token_latency_ms": 13.73160487802546,
      "output_token_source": "usage",
      "output_tokens": 42,
      "planned_context_tokens": 4096,
      "prompt_tokens": 2968,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 1008.3442999966792,
      "ttft_ms": 1008.3450000020093,
      "visible_generation_ms": 562.9950999937137
    },
    {
      "content_chunks": 14,
      "decode_tok_s": 74.46196267841451,
      "e2e_ms": 1561.5993999963393,
      "effective_prefill_tok_s": 3100.5031005130127,
      "generation_ms": 604.3353999993997,
      "mean_inter_token_latency_ms": 13.429675555542214,
      "output_token_source": "usage",
      "output_tokens": 46,
      "planned_context_tokens": 4096,
      "prompt_tokens": 2968,
      "reasoning_chunks": 3,
      "request_index": 2,
      "time_to_first_output_ms": 957.2639999969397,
      "ttft_ms": 1048.4871000007843,
      "visible_generation_ms": 513.1122999955551
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260829T203828Z",
  "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": "2026-08-29T20:38:28Z",
  "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": 9586.086699997395
}