You implement skip_connection function using convolution in ResBlock. But the convolution operation will block the backward gradient. How should I solve this problem? My python version is 3.8.5. Pytorch version is 1.9.0. cuda version is 11.1. The following is a screenshot of the error.
You implement skip_connection function using convolution in ResBlock. But the convolution operation will block the backward gradient. How should I solve this problem? My python version is 3.8.5. Pytorch version is 1.9.0. cuda version is 11.1. The following is a screenshot of the error.