Initial ReGen model release

[pzharrington]

Modulus Pull Request

Description

Adds the ReGen model (https://arxiv.org/abs/2406.16947) training and inference code as an example, including reproducibility for paper figures.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.
• The CHANGELOG.md is up to date with these changes.
• An issue is linked to this pull request.

Dependencies

Additional dependencies for this example specified in requirements.txt

[pzharrington]

/blossom-ci

[nbren12]

LGTM. Would be nice to delete the pdfs though. There are definitely a lot of follow-up focused on usability, but there probably isn't much point to making the example inference usable until we release a checkpoint.

[CharlelieLrt]

Just left a few comments, mostly about:

1. A few binary files. I'm not sure if those are required for the paper submission but it would be better not to have them in the main
2. Some utilities that are already implemented in modulus, which could be re-used if possible.
3. The model that has many poor performance operations. This could re-use many of the layers of the SongUNet defined for CorrDiff.
   I understand it's for paper submission, so feel free to address these later in another PR. Otherwise, LGTM.

[pzharrington]

@CharlelieLrt thanks for the review, I agree with all your comments. As Noah mentioned most of this is in there for direct paper reproducibility/journal submission -- I have a version of the training loop etc in development that uses the physicsnemo SongUNet and other utils, just got stuck in the process of verifying accuracy but we can hopefully pick that up again after this is merged.

Also thanks @nbren12 for the comments -- should be addressed now.

[pzharrington]

/blossom-ci

[pzharrington]

/blossom-ci

[pzharrington]

/blossom-ci

[ktangsali]

/blossom-ci

[pzharrington]

/blossom-ci