swm-dmc-cheetah

LeJEPA world model checkpoint for the DeepMind Control Suite cheetah task, extracted from the galilai-group/stable-worldmodel release artifacts.

Files

File	Description
config.yaml	Training configuration (ViT-tiny encoder, DiT predictor, epoch/LR schedule)
lejepa_weights.ckpt	PyTorch Lightning training checkpoint — use this for inference (no pickle class-path dependencies)
lejepa_epoch_50_object.ckpt	Object checkpoint pickled against private training codebase — requires missing module.* namespace; cannot be loaded with released upstream code

Usage

See CVlization stable_worldmodel_dmc example for a self-contained Docker example that reconstructs and loads the model from lejepa_weights.ckpt.

Source

• Upstream repo: https://github.com/galilai-group/stable-worldmodel
• Original release artifact: dmc_cheetah_lejepa_ckpt.tar.zst

Citation

If you use these artifacts, please cite the original work:

@misc{maes_lelidec2026swm-1,
      title={stable-worldmodel-v1: Reproducible World Modeling Research and Evaluation},
      author={Lucas Maes and Quentin Le Lidec and Dan Haramati and
              Nassim Massaudi and Damien Scieur and Yann LeCun and
              Randall Balestriero},
      year={2026},
      eprint={2602.08968},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2602.08968},
}