[Consistency] Make sure all xxxForSequenceClassification models support problem_type

[NielsRogge]

A while ago (#11012), an additional attribute called problem_type has been added to xxxForSequenceClassification models, which you can set to "multi_label_classification", "single_label_classification" or "regression" to fine-tune xxxForSequenceClassification models for the respective problem. This makes sure the appropriate loss function is used.

This is great, however, 3 things need to be improved in my opinion:

• this is causing a bit of an inconsistency as it's not implemented in all models which have an xxxForSequenceClassification head model.
• it's also not included in any of the CookieCutter templates, which are used to add new models.
• this is not documented anywhere.

cc @abhishekkrthakur

[abhishekkrthakur]

Pretty sure it is documented :)
https://huggingface.co/transformers/main_classes/configuration.html

Agree that its not in cookie-cutter and its not implemented for all models (it cannot be implemented for all). :)

[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.