Add sequence classification capability to Granite models

[jmriosal]

What does this PR do?

Add sequence classification capabilities to the family of Granite models (Granite, GraniteMoe, GraniteMoeHybrid, and GraniteMoeShared).

Fixes #44214, #35720

Why

The Granite models currently only have the base model and causal model heads, so this addition brings them more in line with other models in the library.

Proposed solution and description of changes

The following ForSequenceClassification classes were added:

• GraniteForSequenceClassification
• GraniteMoeForSequenceClassification
• GraniteMoeHybridForSequenceClassification
• GraniteMoeSharedForSequenceClassification

using the existing GenericForSequenceClassification, following the established pattern seen in many other models in the library. Code changes were minimal and done in a way to keep consistent logic across similar models. Changes were implemented in modular_*.py and then modeling_*.py files were automatically generated using utils/modular_model_converter.py

Updated __all__ exports with new classes in each model module.

The Auto Model Registry (MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES) has been updated to allow for auto-loading them via AutoModelForSequenceClassification

New features usage

After with PR, users should be able to load any Granite model variant for sequence classification as follows:

from transformers import AutoModelForSequenceClassification

model = AutoModelForSequenceClassification.from_pretrained(granite-model-id)

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 Add sequence classification capabilities to the Granite models #44214
  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?

Anyone in the community is free to review the PR once the tests have passed.

@ArthurZucker @Cyrilvallez

[jmriosal]

working on adding new tests for the new ForSequenceClassification classes

[jmriosal]

The CI check_repository_consistency is failing due to a pre-existing mismatch in PR #44176 between modular and modeling files for GraniteMoeHybrid, in which modeling_granitemoehybrid.py was modified directly but modular_granitemoehybrid.py was not. This mismatch exists in the main branch and is not caused by these changes.

As a consequence, the following check FAILS:

python utils/check_modular_conversion.py --files src/transformers/models/granitemoehybrid/modular_granitemoehybrid.py

Any advise?

[github-actions]

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

run-slow: auto, granite, granitemoe, granitemoehybrid, granitemoeshared

[ArthurZucker]

LGTM but you can also use GenericForSequenceClassification directly!
Kind of up to you but its generic so compatible!

[ArthurZucker]

don't worry we'll fix this one on main!