Hello there,
for evaluation purposes I ran 2 extra losses alongside the first main loss.
I only called backward on the main loss.
Do the other 2 loss calculations change anything for back propagation or any other update?
Greetings
Lemling
Hello there,
for evaluation purposes I ran 2 extra losses alongside the first main loss.
I only called backward on the main loss.
Do the other 2 loss calculations change anything for back propagation or any other update?
Greetings
Lemling
unless you call .backward()
on the two extra losses, or add them to the main loss before calling backward (for eg. `(main_loss + loss_extra1 + loss_extra2).backward()), they wont affect the gradients you computed.