TODO List Support Python3 Speed up using Pytorch's native dimension format Try parallelising over a larger number of similar GPUs for a decrease in epoch time