Teacher forcing option with PackedSequence input


Can a teacher forcing option (i.e. using predicted vs. ground-truth label as current input during training for each time step in a sequence prediction) be controlled when using a PackedSequence object as input to an RNN? It seems like teacher forcing is performed by default in such a case. I am not sure what the process would be like as there are no code examples I could find online. Examples I did find manually iterate through each time step of a sequence instead.

See my reply to your other post.