{
  "base_url": "http://127.0.0.1:39118/v1",
  "completed": 4,
  "concurrency": 4,
  "context_distribution": {
    "120000": 4
  },
  "context_seed": 0,
  "context_tokens": 120000,
  "engine": "sglang-glm53-sm120-v0.1.1-rc14",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-02T10:49:18Z",
  "gpu": "2x RTX PRO 6000 Blackwell Max-Q TP2 WSL2",
  "manifest": null,
  "max_context_tokens": 499712,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 80,
    "content_chunks_s": null,
    "decode_tok_s_p50": 0.34682144660430075,
    "decode_tok_s_p95": 90.18851577225323,
    "e2e_p50_ms": 298796.0312999785,
    "e2e_p95_ms": 299106.44609999144,
    "effective_prefill_tok_s_p50": 434.3036826062556,
    "effective_prefill_tok_s_p95": 1261.2275044787066,
    "generation_p50_ms": 782.5730000040494,
    "generation_p95_ms": 222522.59260002756,
    "mean_inter_token_latency_ms_p50": 17.01245652182716,
    "mean_inter_token_latency_ms_p95": 4363.188090196619,
    "output_token_sources": {
      "usage": 4
    },
    "output_tokens": 182,
    "prefix_cache_hit_avg": 0.030577641145982094,
    "prompt_token_samples": 4,
    "prompt_tokens": 385115,
    "prompt_tokens_p50": 96278.0,
    "prompt_tokens_p95": 96281.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 0.6084736354069853,
    "time_to_first_output_p50_ms": 148862.97349998495,
    "time_to_first_output_p95_ms": 298784.89740000805,
    "ttft_p50_ms": 148862.973899988,
    "ttft_p95_ms": 298784.8978000111,
    "ttft_samples": 4
  },
  "model": "glm53-flash-ormandj-sglang-sm120-tp2-499k-adaptive-mtp",
  "request_timings": [
    {
      "content_chunks": 12,
      "decode_tok_s": 90.18851577225323,
      "e2e_ms": 299106.44609999144,
      "effective_prefill_tok_s": 322.2251219448597,
      "generation_ms": 321.54869998339564,
      "mean_inter_token_latency_ms": 11.087886206323988,
      "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": 298784.89740000805,
      "ttft_ms": 298784.8978000111,
      "visible_generation_ms": 321.54829998034984
    },
    {
      "content_chunks": 24,
      "decode_tok_s": 0.2291902112234948,
      "e2e_ms": 298859.3356999918,
      "effective_prefill_tok_s": 1261.2275044787066,
      "generation_ms": 222522.59260002756,
      "mean_inter_token_latency_ms": 4363.188090196619,
      "output_token_source": "usage",
      "output_tokens": 52,
      "planned_context_tokens": 120000,
      "prompt_tokens": 96278,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 76336.74309996422,
      "ttft_ms": 76336.74369996879,
      "visible_generation_ms": 222522.592000023
    },
    {
      "content_chunks": 20,
      "decode_tok_s": 58.78045881951201,
      "e2e_ms": 222470.76920000836,
      "effective_prefill_tok_s": 434.3036826062556,
      "generation_ms": 782.5730000040494,
      "mean_inter_token_latency_ms": 17.01245652182716,
      "output_token_source": "usage",
      "output_tokens": 47,
      "planned_context_tokens": 120000,
      "prompt_tokens": 96280,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 221688.1962000043,
      "ttft_ms": 221688.19640000584,
      "visible_generation_ms": 782.5728000025265
    },
    {
      "content_chunks": 24,
      "decode_tok_s": 0.34682144660430075,
      "e2e_ms": 298796.0312999785,
      "effective_prefill_tok_s": 646.776009751073,
      "generation_ms": 149933.05779999355,
      "mean_inter_token_latency_ms": 2883.3280346152605,
      "output_token_source": "usage",
      "output_tokens": 53,
      "planned_context_tokens": 120000,
      "prompt_tokens": 96281,
      "reasoning_chunks": 0,
      "request_index": 3,
      "time_to_first_output_ms": 148862.97349998495,
      "ttft_ms": 148862.973899988,
      "visible_generation_ms": 149933.0573999905
    }
  ],
  "requests": 4,
  "run_id": "benchmark-20260902T104419Z",
  "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": "2026-09-02T10:44:19Z",
  "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": 299109.0976000414
}