Enable automatic creation of decoder_input_ids given labels in TF

[Carlosbogo]

Added a way to automatically create the decoder_input_ids from the labels provided by the user the TFEncoderDecoderModel

[LysandreJik]

Hey @Carlosbogo, thanks for your contribution! I've pinged @patrickvonplaten to review your PR, but please note that Patrick if off until next Monday. Thanks for your patience!

[patrickvonplaten]

Thanks for the PR @Carlosbogo!

I think we should also move the loss computation into the TFEncoderDecoderModel class similar to how @NielsRogge has done it in this PR: #14139 . Right now the loss is computed in the decoder. However we should compute the loss directly in the TFEncoderDecoderModel class.

[NielsRogge]

Hi @Carlosbogo, thanks for your contribution! Let me know if you are able to take into account the comment above.

Thanks!

[Carlosbogo]

Yeah, I think I can work on that.
Should I create a new PR for it when I finish?

[NielsRogge]

No it can be included in this PR.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[patrickvonplaten]

Hey @Carlosbogo,

Hope you had a good start to 2022! Let me know if I can help with anything to get this PR merged :-)

[Carlosbogo]

I'm really sorry for the delay: I had my university exams and couldn't find time to work on it. I'll try to do it as soon as possible.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.