{
  "base_url": "http://127.0.0.1:39068/v1",
  "completed": 0,
  "concurrency": 1,
  "context_distribution": {
    "70000": 1
  },
  "context_seed": 0,
  "context_tokens": 70000,
  "engine": "vllm-r16-b12x-wsl2-offload-mmap-unpinned",
  "failed": 1,
  "failures": [
    {
      "error_message": "stream completed without visible content",
      "error_message_truncated": false,
      "error_type": "ValueError",
      "request_index": 0
    }
  ],
  "finished_at": "2026-08-02T06:06:42Z",
  "gpu": "2x RTX PRO 6000 Blackwell Max-Q TP2",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 256,
  "measurement_protocol": "capacity-v3",
  "metrics": {
    "content_chunks": 0,
    "content_chunks_s": null,
    "decode_tok_s_p50": null,
    "decode_tok_s_p95": null,
    "e2e_p50_ms": null,
    "e2e_p95_ms": null,
    "effective_prefill_tok_s_p50": null,
    "effective_prefill_tok_s_p95": null,
    "generation_p50_ms": null,
    "generation_p95_ms": null,
    "mean_inter_token_latency_ms_p50": null,
    "mean_inter_token_latency_ms_p95": null,
    "output_token_sources": {},
    "output_tokens": 0,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 0,
    "prompt_tokens": null,
    "prompt_tokens_p50": null,
    "prompt_tokens_p95": null,
    "reasoning_chunks": 0,
    "throughput_tok_s": 0.0,
    "time_to_first_output_p50_ms": null,
    "time_to_first_output_p95_ms": null,
    "ttft_p50_ms": null,
    "ttft_p95_ms": null,
    "ttft_samples": 0
  },
  "model": "deepseek-v4-flash-0731-r16-b12x-dspark5-offload8-wsl2-mmap-unpinned-tp2-256k",
  "request_timings": [],
  "requests": 1,
  "run_id": "benchmark-20260802T060632Z",
  "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": "2026-08-02T06:06:32Z",
  "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": 9839.17229999497
}