Hey! Is it not possible to put operations in the "on_forward" hook that are incorporated into the back-propagation? E.g. I might do some operations in the "on_forward"-hook that I assign to self.task.key that I later use in the loss and that I want to part of the back-propagation. If so, where can I see this in the code?
Hey! Is it not possible to put operations in the "on_forward" hook that are incorporated into the back-propagation? E.g. I might do some operations in the "on_forward"-hook that I assign to self.task.key that I later use in the loss and that I want to part of the back-propagation. If so, where can I see this in the code?