[Abdennacer-Badaoui]

This PR fixes a bug in ProphetNet where passing encoder_outputs as a tuple to forward() would raise:

AttributeError: 'tuple' object has no attribute 'last_hidden_state'

This happens because the forward() method assumes encoder_outputs is a BaseModelOutput when provided manually, while the docstring says it can be a tuple.

Changes:

Added a check in forward():

Convert tuple encoder_outputs to BaseModelOutput format before passing to ProphetNetSeq2SeqModelOutput.

if isinstance(encoder_outputs, tuple):
    encoder_outputs = BaseModelOutput(
        last_hidden_state=encoder_outputs[0],
        hidden_states=encoder_outputs[1] if len(encoder_outputs) > 1 else None,
        attentions=encoder_outputs[2] if len(encoder_outputs) > 2 else None,
    )

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[Rocketknight1]

cc @ArthurZucker for text models

[ArthurZucker]

I think we'd rather update the docstring as it should no longer be a tuple!

[Abdennacer-Badaoui]

Alright! In this case, the ProphetNetEncoder should be modified to always return a BaseModelOutput. This means removing the return_dict argument, as it will not be used there.

[ArthurZucker]

btw if you have time, can you add the check_model_input decorator, updating to the latest standards? 🤗

[ArthurZucker]

sorry, what I mean by that is add this, remove the args, add the TransformersKwargs, remove the

        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )

etc! Have a look at modeling llama to see how we handle ti

[Abdennacer-Badaoui]

My bad, it’s clearer now. I’ll try to fix it. Thanks!

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: prophetnet

[ArthurZucker]

Thanks for the PR, and sorry it didn't make it in! This branch is based on v4, and v5 is very different, so rather than rebasing we'd love to invite you to open a new PR against main using our latest tools and good practices:

• Transformers tenets
• mlinter and the modeling rules it enforces (make typing)
• Modular transformers
• Contributing guide

Closing this one in the meantime, thanks again!