anyone getting this issue running multiple GPUs?
Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cuda:1! (when checking argument for argument index in method wrapper_CUDA__index_select)