Hi there, I am just wondering that in Pytorch, if I do:
loss=loss/loss.item(), I know that the resulting loss will be having a scalar value of 1. Do the gradients of the loss still preserve? What is the difference between loss/loss and loss/loss.item() ?