Hi @hovren, There are two scenarios that are relevant during training: 1. Your gradients need to be reset for every batch after back propagation. ...