{
  "base_url": "http://127.0.0.1:39118/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "120000": 3
  },
  "context_seed": 0,
  "context_tokens": 120000,
  "engine": "sglang-glm53-sm120-v0.1.1-rc14",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-02T11:35:45Z",
  "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": 55,
    "content_chunks_s": null,
    "decode_tok_s_p50": 93.3469984676811,
    "decode_tok_s_p95": 105.6658913554335,
    "e2e_p50_ms": 17170.852999959607,
    "e2e_p95_ms": 17227.093400026206,
    "effective_prefill_tok_s_p50": 5762.69030860975,
    "effective_prefill_tok_s_p95": 5952.868462595826,
    "generation_p50_ms": 463.72579998569563,
    "generation_p95_ms": 504.30580001557246,
    "mean_inter_token_latency_ms_p50": 10.712717242282013,
    "mean_inter_token_latency_ms_p95": 10.729910638629201,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 128,
    "prefix_cache_hit_avg": 0.03057875275250322,
    "prompt_token_samples": 3,
    "prompt_tokens": 288834,
    "prompt_tokens_p50": 96278.0,
    "prompt_tokens_p95": 96280.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 2.515345024443562,
    "time_to_first_output_p50_ms": 16707.12719997391,
    "time_to_first_output_p95_ms": 16722.787600010633,
    "ttft_p50_ms": 16707.12769997772,
    "ttft_p95_ms": 16722.78800001368,
    "ttft_samples": 3
  },
  "model": "glm53-flash-ormandj-sglang-sm120-tp2-240k-c1-adaptive-mtp",
  "request_timings": [
    {
      "content_chunks": 13,
      "decode_tok_s": 93.3469984676811,
      "e2e_ms": 16483.71220001718,
      "effective_prefill_tok_s": 5952.868462595826,
      "generation_ms": 310.66880002617836,
      "mean_inter_token_latency_ms": 10.712717242282013,
      "output_token_source": "usage",
      "output_tokens": 30,
      "planned_context_tokens": 120000,
      "prompt_tokens": 96276,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 16173.043399991002,
      "ttft_ms": 16173.044099996332,
      "visible_generation_ms": 310.6681000208482
    },
    {
      "content_chunks": 20,
      "decode_tok_s": 105.6658913554335,
      "e2e_ms": 17170.852999959607,
      "effective_prefill_tok_s": 5762.69030860975,
      "generation_ms": 463.72579998569563,
      "mean_inter_token_latency_ms": 9.463791836442768,
      "output_token_source": "usage",
      "output_tokens": 50,
      "planned_context_tokens": 120000,
      "prompt_tokens": 96278,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 16707.12719997391,
      "ttft_ms": 16707.12769997772,
      "visible_generation_ms": 463.7252999818884
    },
    {
      "content_chunks": 22,
      "decode_tok_s": 93.19742108567597,
      "e2e_ms": 17227.093400026206,
      "effective_prefill_tok_s": 5757.413315465346,
      "generation_ms": 504.30580001557246,
      "mean_inter_token_latency_ms": 10.729910638629201,
      "output_token_source": "usage",
      "output_tokens": 48,
      "planned_context_tokens": 120000,
      "prompt_tokens": 96280,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 16722.787600010633,
      "ttft_ms": 16722.78800001368,
      "visible_generation_ms": 504.30540001252666
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260902T113455Z",
  "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": "2026-09-02T11:34:55Z",
  "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": 50887.65109999804
}