Run language models locally without NVIDIA.
Recent revisions are kept. Restoring one creates a new revision.
