Fix problem_type to match with the applied loss function for distillbert sequence classification

[sidhantls]

What does this PR do?

The problem_type in config is not correct with the loss function applied. Can be seen here.

This PR fixes this so that the applied loss is consistent with the problem type: BCEWithLogitsLoss is applied for the problem_type of single_label_classification, and CrossEntropyLoss is applied for the problem type multi_label_classification

Fixes #12014

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?

@LysandreJik

[LysandreJik]

cc @abhi1thakur @sgugger

[sgugger]

No, this is incorrect. What single_label_classification means each sample can only have one label (but there could be multiple classes) so the loss to use is cross entropy. multi_label_classification means each sample can have zero or several labels, so in this case we use bce (because there can't be a softmax).