saev - Sparse Auto-Encoders for Vision

Training sparse autoencoders (SAEs) on vision transformers (ViTs), implemented in PyTorch.

Docs

• Docs
• Quickstart
• User guide
• API reference

Research

This framework was developed for a series of projects leveraging SAEs with vision models.

• [Feb 2025] Interpretable and Testable Vision Features via Sparse Autoencoders
• [Nov 2025] Towards Open-Ended Visual Scientific Discovery with Sparse Autoencoders

Trained SAE checkpoints are available at:

• Huggingface Models

If you want to cite the software, please cite it as:

@software{stevens2025saev,
  author = {Stevens, Samuel},
  license = {CC-BY-4.0},
  month = apr,
  title = {{saev}},
  url = {https://github.com/OSU-NLP-Group/saev},
  year = {2025}
}