{
  "base_url": "http://127.0.0.1:8001/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "500000": 3
  },
  "context_seed": 0,
  "context_tokens": 500000,
  "engine": "vllm-glm53-exl3-k3-b12x-dflash2-k5",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-08-29T23:40:31Z",
  "gpu": "dual-rtx-pro-6000-max-q-tp2",
  "manifest": null,
  "max_context_tokens": 1048576,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 45,
    "content_chunks_s": null,
    "decode_tok_s_p50": 61.41669626201493,
    "decode_tok_s_p95": 76.78278183696254,
    "e2e_p50_ms": 123972.87219999998,
    "e2e_p95_ms": 132715.58490000461,
    "effective_prefill_tok_s_p50": 3202.963514917107,
    "effective_prefill_tok_s_p95": 3213.465327192788,
    "generation_p50_ms": 612.1163999996497,
    "generation_p95_ms": 633.3058000018355,
    "mean_inter_token_latency_ms_p50": 16.282217391404707,
    "mean_inter_token_latency_ms_p95": 16.66594210531146,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 111,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 3,
    "prompt_tokens": 1185355,
    "prompt_tokens_p50": 395118.0,
    "prompt_tokens_p95": 395120.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 0.2918742907421898,
    "time_to_first_output_p50_ms": 123360.75580000033,
    "time_to_first_output_p95_ms": 132341.0939000023,
    "ttft_p50_ms": 123360.75630000414,
    "ttft_p95_ms": 132341.09480000188,
    "ttft_samples": 3
  },
  "model": "glm53-flash-exl3-k3-dflash2-k5-fp8-tp2-1m-vision",
  "request_timings": [
    {
      "content_chunks": 10,
      "decode_tok_s": 61.41669626201493,
      "e2e_ms": 132715.58490000461,
      "effective_prefill_tok_s": 2985.5956933418784,
      "generation_ms": 374.4910000023083,
      "mean_inter_token_latency_ms": 16.282217391404707,
      "output_token_source": "usage",
      "output_tokens": 24,
      "planned_context_tokens": 500000,
      "prompt_tokens": 395117,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 132341.0939000023,
      "ttft_ms": 132341.09480000188,
      "visible_generation_ms": 374.4901000027312
    },
    {
      "content_chunks": 16,
      "decode_tok_s": 60.002608534281336,
      "e2e_ms": 123590.2883000017,
      "effective_prefill_tok_s": 3213.465327192788,
      "generation_ms": 633.3058000018355,
      "mean_inter_token_latency_ms": 16.66594210531146,
      "output_token_source": "usage",
      "output_tokens": 39,
      "planned_context_tokens": 500000,
      "prompt_tokens": 395118,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 122956.98249999987,
      "ttft_ms": 122956.98339999944,
      "visible_generation_ms": 633.3049000022584
    },
    {
      "content_chunks": 19,
      "decode_tok_s": 76.78278183696254,
      "e2e_ms": 123972.87219999998,
      "effective_prefill_tok_s": 3202.963514917107,
      "generation_ms": 612.1163999996497,
      "mean_inter_token_latency_ms": 13.023753191481909,
      "output_token_source": "usage",
      "output_tokens": 48,
      "planned_context_tokens": 500000,
      "prompt_tokens": 395120,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 123360.75580000033,
      "ttft_ms": 123360.75630000414,
      "visible_generation_ms": 612.1158999958425
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260829T233411Z",
  "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": "2026-08-29T23:34:11Z",
  "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": 380300.7099999959
}