quantization.json

{
  "method": "fp8_e4m3_per_channel_scale",
  "dtype": "float8_e4m3fn",
  "scale_dtype": "float32",
  "num_tensors": 338
}