{
  "base_url": "http://127.0.0.1:39118/v1",
  "completed": 3,
  "concurrency": 1,
  "context_distribution": {
    "380000": 3
  },
  "context_seed": 0,
  "context_tokens": 380000,
  "engine": "sglang-glm53-sm120-v0.1.1-rc14",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-02T11:17:12Z",
  "gpu": "2x RTX PRO 6000 Blackwell Max-Q TP2 WSL2",
  "manifest": null,
  "max_context_tokens": 393216,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 62,
    "content_chunks_s": null,
    "decode_tok_s_p50": 99.79115138353907,
    "decode_tok_s_p95": 106.3145881935987,
    "e2e_p50_ms": 56362.32839996228,
    "e2e_p95_ms": 56454.593600006774,
    "effective_prefill_tok_s_p50": 5456.732792727141,
    "effective_prefill_tok_s_p95": 5470.288001990131,
    "generation_p50_ms": 561.1719999578781,
    "generation_p95_ms": 568.6509999795817,
    "mean_inter_token_latency_ms_p50": 10.020928570676395,
    "mean_inter_token_latency_ms_p95": 11.373019999591634,
    "output_token_sources": {
      "usage": 3
    },
    "output_tokens": 141,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 3,
    "prompt_tokens": 913473,
    "prompt_tokens_p50": 304491.0,
    "prompt_tokens_p95": 304492.0,
    "reasoning_chunks": 0,
    "throughput_tok_s": 0.8353306078293593,
    "time_to_first_output_p50_ms": 55801.1564000044,
    "time_to_first_output_p95_ms": 55885.94260002719,
    "ttft_p50_ms": 55801.15680000745,
    "ttft_p95_ms": 55885.943100031,
    "ttft_samples": 3
  },
  "model": "glm53-flash-ormandj-sglang-sm120-tp2-393k-c1-adaptive-mtp",
  "request_timings": [
    {
      "content_chunks": 13,
      "decode_tok_s": 106.3145881935987,
      "e2e_ms": 55963.51070003584,
      "effective_prefill_tok_s": 5470.288001990131,
      "generation_ms": 300.9934999863617,
      "mean_inter_token_latency_ms": 9.406046874573804,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 380000,
      "prompt_tokens": 304490,
      "reasoning_chunks": 0,
      "request_index": 0,
      "time_to_first_output_ms": 55662.51720004948,
      "ttft_ms": 55662.518199998885,
      "visible_generation_ms": 300.9925000369549
    },
    {
      "content_chunks": 24,
      "decode_tok_s": 87.92739307905082,
      "e2e_ms": 56454.593600006774,
      "effective_prefill_tok_s": 5448.436330030727,
      "generation_ms": 568.6509999795817,
      "mean_inter_token_latency_ms": 11.373019999591634,
      "output_token_source": "usage",
      "output_tokens": 51,
      "planned_context_tokens": 380000,
      "prompt_tokens": 304491,
      "reasoning_chunks": 0,
      "request_index": 1,
      "time_to_first_output_ms": 55885.94260002719,
      "ttft_ms": 55885.943100031,
      "visible_generation_ms": 568.6504999757744
    },
    {
      "content_chunks": 25,
      "decode_tok_s": 99.79115138353907,
      "e2e_ms": 56362.32839996228,
      "effective_prefill_tok_s": 5456.732792727141,
      "generation_ms": 561.1719999578781,
      "mean_inter_token_latency_ms": 10.020928570676395,
      "output_token_source": "usage",
      "output_tokens": 57,
      "planned_context_tokens": 380000,
      "prompt_tokens": 304492,
      "reasoning_chunks": 0,
      "request_index": 2,
      "time_to_first_output_ms": 55801.1564000044,
      "ttft_ms": 55801.15680000745,
      "visible_generation_ms": 561.1715999548323
    }
  ],
  "requests": 3,
  "run_id": "benchmark-20260902T111423Z",
  "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": "2026-09-02T11:14:23Z",
  "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": 168795.44299998088
}