{
  "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-vision",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-29T20:40:31Z",
  "gpu": "dual-rtx-pro-6000-tp2",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 57,
    "content_chunks_s": null,
    "decode_tok_s_p50": 55.70977130148541,
    "decode_tok_s_p95": 64.82724658558394,
    "e2e_p50_ms": 33687.94049999997,
    "e2e_p95_ms": 39647.311900000204,
    "effective_prefill_tok_s_p50": 3138.4734906600515,
    "effective_prefill_tok_s_p95": 3155.7105644011435,
    "generation_p50_ms": 839.6651000002748,
    "generation_p95_ms": 848.4086999960709,
    "mean_inter_token_latency_ms_p50": 17.950172413889206,
    "mean_inter_token_latency_ms_p95": 19.083297727278975,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 131,
    "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": 1.2261118396903128,
    "time_to_first_output_p50_ms": 32839.5318000039,
    "time_to_first_output_p95_ms": 39126.75689999742,
    "ttft_p50_ms": 32839.53219999967,
    "ttft_p95_ms": 39126.75840000156,
    "ttft_samples": 3
  },
  "model": "glm53-flash-tr3-4bpw-tp2-262k-fixed-mtp5-vision",
  "request_timings": [
    {
      "content_chunks": 13,
      "decode_tok_s": 55.70977130148541,
      "e2e_ms": 39647.311900000204,
      "effective_prefill_tok_s": 2634.156474134119,
      "generation_ms": 520.555000002787,
      "mean_inter_token_latency_ms": 17.950172413889206,
      "output_token_source": "usage",
      "output_tokens": 30,
      "planned_context_tokens": 128000,
      "prompt_tokens": 103066,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 39126.75689999742,
      "ttft_ms": 39126.75840000156,
      "visible_generation_ms": 520.5534999986412
    },
    {
      "content_chunks": 22,
      "decode_tok_s": 52.401844497271114,
      "e2e_ms": 33500.13819999731,
      "effective_prefill_tok_s": 3155.7105644011435,
      "generation_ms": 839.6651000002748,
      "mean_inter_token_latency_ms": 19.083297727278975,
      "output_token_source": "usage",
      "output_tokens": 45,
      "planned_context_tokens": 128000,
      "prompt_tokens": 103067,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 32660.473099997034,
      "ttft_ms": 32660.47399999661,
      "visible_generation_ms": 839.6642000006977
    },
    {
      "content_chunks": 22,
      "decode_tok_s": 64.82724658558394,
      "e2e_ms": 33687.94049999997,
      "effective_prefill_tok_s": 3138.4734906600515,
      "generation_ms": 848.4086999960709,
      "mean_inter_token_latency_ms": 15.42561272720129,
      "output_token_source": "usage",
      "output_tokens": 56,
      "planned_context_tokens": 128000,
      "prompt_tokens": 103066,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 32839.5318000039,
      "ttft_ms": 32839.53219999967,
      "visible_generation_ms": 848.4083000003011
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260829T203845Z",
  "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": "2026-08-29T20:38:45Z",
  "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": 106841.80330000527
}