Issue with pytorch tensors and multiple GPUs when using DataParallel...

Parameters can't be updated when using torch.nn.DataParallel to train on multiple GPUs...