On certain ROCm devices, when using float16 inputs this module will use different precision for backward. Parameters: in_features (int) – size of each input ...
確定! 回上一頁