{
  "base_url": "http://127.0.0.1:39118/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "230000": 3
  },
  "context_seed": 0,
  "context_tokens": 230000,
  "engine": "sglang-glm53-sm120-v0.1.1-rc14",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-02T11:37:35Z",
  "gpu": "2x RTX PRO 6000 Blackwell Max-Q TP2 WSL2",
  "manifest": null,
  "max_context_tokens": 245760,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 51,
    "content_chunks_s": null,
    "decode_tok_s_p50": 95.0049206970752,
    "decode_tok_s_p95": 118.68974434805479,
    "e2e_p50_ms": 34225.277699995786,
    "e2e_p95_ms": 34281.407200032845,
    "effective_prefill_tok_s_p50": 5607.741676837262,
    "effective_prefill_tok_s_p95": 5627.907447602635,
    "generation_p50_ms": 357.8761999960989,
    "generation_p95_ms": 466.1905000102706,
    "mean_inter_token_latency_ms_p50": 10.525770588120555,
    "mean_inter_token_latency_ms_p95": 11.370500000250503,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 114,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 3,
    "prompt_tokens": 568880,
    "prompt_tokens_p50": 189627.0,
    "prompt_tokens_p95": 189627.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 1.112044730106607,
    "time_to_first_output_p50_ms": 33815.216700022575,
    "time_to_first_output_p95_ms": 33867.40149999969,
    "ttft_p50_ms": 33815.21720002638,
    "ttft_p95_ms": 33867.40219994681,
    "ttft_samples": 3
  },
  "model": "glm53-flash-ormandj-sglang-sm120-tp2-240k-c1-adaptive-mtp",
  "request_timings": [
    {
      "content_chunks": 16,
      "decode_tok_s": 95.0049206970752,
      "e2e_ms": 34225.277699995786,
      "effective_prefill_tok_s": 5599.071425659915,
      "generation_ms": 357.8761999960989,
      "mean_inter_token_latency_ms": 10.525770588120555,
      "output_token_source": "usage",
      "output_tokens": 35,
      "planned_context_tokens": 230000,
      "prompt_tokens": 189626,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 33867.40149999969,
      "ttft_ms": 33867.40219994681,
      "visible_generation_ms": 357.8755000489764
    },
    {
      "content_chunks": 14,
      "decode_tok_s": 118.68974434805479,
      "e2e_ms": 33997.36269999994,
      "effective_prefill_tok_s": 5627.907447602635,
      "generation_ms": 303.31180000212044,
      "mean_inter_token_latency_ms": 8.425327777836678,
      "output_token_source": "usage",
      "output_tokens": 37,
      "planned_context_tokens": 230000,
      "prompt_tokens": 189627,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 33694.05089999782,
      "ttft_ms": 33694.05140000163,
      "visible_generation_ms": 303.3112999983132
    },
    {
      "content_chunks": 21,
      "decode_tok_s": 87.94688008249145,
      "e2e_ms": 34281.407200032845,
      "effective_prefill_tok_s": 5607.741676837262,
      "generation_ms": 466.1905000102706,
      "mean_inter_token_latency_ms": 11.370500000250503,
      "output_token_source": "usage",
      "output_tokens": 42,
      "planned_context_tokens": 230000,
      "prompt_tokens": 189627,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 33815.216700022575,
      "ttft_ms": 33815.21720002638,
      "visible_generation_ms": 466.1900000064634
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260902T113552Z",
  "schema": "anvil-serving.benchmark/v1",
  "serve_flags": {
    "no_thinking": false,
    "reasoning_effort": null,
    "shared_prefix_burst": false,
    "thinking_mode": "disabled"
  },
  "source_recipe": null,
  "started_at": "2026-09-02T11:35:52Z",
  "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": 102513.86200002162
}