Add supports_gradient_checkpointing to NemotronHPreTrainedModel

[sergiopaniego]

What does this PR do?

Enables gradient checkpointing for NemotronH by setting supports_gradient_checkpointing = True on NemotronHPreTrainedModel. The idea comes from its usage in TRL.

NemotronHBlock already inherits from GradientCheckpointingLayer, so the infrastructure at the block level is in place. The only missing piece was the class-level flag, which currently defaults to False (inherited from PreTrainedModel). As a result, any call to model.gradient_checkpointing_enable() (including the one issued by Trainer when gradient_checkpointing=True) raises:

ValueError: NemotronHForCausalLM does not support gradient checkpointing.

This is a simple omission, not a limitation of the architecture. All sibling hybrid Mamba/attention models in the library already enable it:

Model	GradientCheckpointingLayer	supports_gradient_checkpointing
Bamba	✅	✅
GraniteMoeHybrid	✅	✅
Zamba2	✅	✅
NemotronH	✅	❌ (before this PR)

GraniteMoeHybrid is the closest analogue (MoE + hybrid Mamba/attention, same layout as NemotronH).

The change is made in modular_nemotron_h.py and propagated to modeling_nemotron_h.py via utils/modular_model_converter.py.

Fixes the failure seen downstream in huggingface/trl#5278, where NemotronH tests required a gradient_checkpointing=False workaround.

Fixes # (issue)

Code Agent Policy

The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by
code agents. We are currently bottlenecked by our ability to review and respond to them. As a result,
we ask that new users do not submit pure code agent PRs at this time.
You may use code agents in drafting or to help you diagnose issues. We'd also ask autonomous "OpenClaw"-like agents
not to open any PRs or issues for the moment.

PRs that appear to be fully agent-written will probably be closed without review, and we may block users who do this
repeatedly or maliciously.

This is a rapidly-evolving situation that's causing significant shockwaves in the open-source community. As a result,
this policy is likely to be updated regularly in the near future. For more information, please read CONTRIBUTING.md.

• I confirm that this is not a pure code agent PR.

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? (n/a — class attribute only)
• Did you write any new necessary tests? (n/a — covered by existing hybrid-model GC tests)

Who can review?

@ArthurZucker @Cyrilvallez

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[github-actions]

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

run-slow: nemotron_h

[ArthurZucker]

TY