{
  "probe": "greedy_generation",
  "provider": "CUDAExecutionProvider",
  "session_load_seconds": 28.961127704940736,
  "step_seconds": [
    0.18181518698111176,
    0.030539471190422773,
    0.028461792040616274,
    0.028141363989561796,
    0.028162146918475628,
    0.028294692980125546,
    0.028191918972879648,
    0.028165633091703057
  ],
  "finite_logits": true,
  "versions": {
    "python": "3.12.9",
    "onnxruntime": "1.29.0",
    "onnx_ir": "1.0.0",
    "transformers": "5.15.1",
    "torch": "2.13.0+cu130",
    "numpy": "2.5.2",
    "providers_available": [
      "TensorrtExecutionProvider",
      "CUDAExecutionProvider",
      "CPUExecutionProvider"
    ]
  },
  "hardware": "NVIDIA H200 host; provider is recorded per probe"
}