I want to measure peak memory usage in TensorFlow 1.13 per epoch.
In PyTorch there is a very simple solution by using combination of two functions: torch.cuda.m
torch.cuda.m