Fix failing tests for XLA generation in TF

[dsuess]

What does this PR do?

This PR fixes the failing tests for text generation in XLA mentioned in #17935. The issue was that the failing models had additional settings limiting the sequence length, which weren’t updated when creating the config-objects for the tests.

Fixes # 17935

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?

@gante @LysandreJik

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[gante]

Hi @dsuess! Thank you so much for looking into this -- I can confirm that the tests are passing after these changes.

This implies two things:

1. The models are indeed XLA-ready, so I will tick their boxes in the original issue TF: XLA generation not working properly in some models #17935
2. The tests were failing due to poor model parametrization at test time -- I will open a follow-up PR to fix them

[gante]

cc @sgugger (with these changes the XLA tests will run instead of being skipped, but the root cause still needs to be addressed to avoid this ad hoc logic, see my comment above)

[sgugger]

Thanks for fixing!