Oh, thank for the quick response, it is very helpful. I still ask for an example having multiple loss. For example
loss = loss1 + loss2+ …
loss.backward()
Will it be a solution? I mean, Will mxnet compute the gradient for loss1, loss2 by only calling loss.backward? or I need to do loss1.backward(), loss2.backward()… for every loss my model has.
Sorry, I know this could be trivial, but I don’t have experience with autograd.