About using dataparrallel

[czb2133]

Thanks for sharing the code.
I wanted to apply the code on multiple GPUs so I used torch.nn.dataparrallel(). However, I found it really hard to change the code, for you used F.conv2d() which accept tensor as paramaters and I need to duplicate it on all the GPUs. Then how to get the grad of the input paramaters and how to ensure all the input paramaters sharing memory?