MultiProcessing with PyTorch

nn.parallel.DataParallel
https://discuss.pytorch.org/t/is-the-loss-function-paralleled-when-using-dataparallel/3346