Load models, run inference, and fine-tune with Transformers.
Recent revisions are kept. Restoring one creates a new revision.