{
  "scope": "warm inference including tensor assembly and device copies; excludes tokenization/network",
  "device": "cpu",
  "prefix_sharing": false,
  "autoregressive_decode_steps": 0,
  "results": [
    {
      "states": 1,
      "questions": 1,
      "candidate_paths": 4,
      "median_ms": 696.8240409623832,
      "questions_per_second": 1.4350825189941794,
      "repeats": 6
    },
    {
      "states": 4,
      "questions": 4,
      "candidate_paths": 16,
      "median_ms": 4200.025249971077,
      "questions_per_second": 0.9523752267983497,
      "repeats": 6
    },
    {
      "states": 8,
      "questions": 8,
      "candidate_paths": 32,
      "median_ms": 8303.48570807837,
      "questions_per_second": 0.9634508062339275,
      "repeats": 6
    }
  ]
}