Do not warn about unexpected decoder weights when loading T5EncoderModel and LongT5EncoderModel

[fleonce]

What does this PR do?

Adds [r"decoder"] to both T5EncoderModel and LongT5EncoderModel, as both models do not have any decoder layers and loading pretrained model checkpoints like t5-small will give warnings about keys found in the checkpoint but not in the model itself. To prevent this issue, r"decoder" has been added to _keys_to_ignore_on_load_unexpected for both model classes.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

@ArthurZucker @younesbelkada

[younesbelkada]

Thanks a lot for this PR!
The explanation makes sense to me! However I was not able to repro on the main branch:

>>> import transformers
>>> model = transformers.T5EncoderModel.from_pretrained("t5-small")
>>> 

Can you share a reproducible snippet?

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

[fleonce]

Thank you for the follow-up question!
The problem can be reproduced when the log level for the whole library has been set to INFO:

import transformers
transformers.logging.set_verbosity_info()
m = transformers.T5EncoderModel.from_pretrained('t5-small')

I was not using the latest version of transformers when I was encountering this issue initially, now it is hidden by default because verbosity info is required to show those kinds of warnings, unless T5EncoderModel is contained in model.config.architectures, but only T5ForConditionalGeneration is in there. This seems to be caused by commit 096f2cf (

transformers/src/transformers/modeling_utils.py

Line 3674 in bc7ce18

warner = logger.warning if model.__class__.__name__ in archs else logger.info

)

The problem persists however, but we wont see the warning by default

[younesbelkada]

Managed to reproduce, this makes sense, thanks !

[ArthurZucker]

thanks!