Citation¶

If you use mlxterp in your research, please cite it as follows:

BibTeX¶

@software{mlxterp2025,
  author = {Sigurd Schacht},
  title = {mlxterp: Mechanistic Interpretability for MLX},
  year = {2025},
  url = {https://github.com/coairesearch/mlxterp},
  version = {0.1.0}
}

APA¶

Schacht, S. (2025). mlxterp: Mechanistic Interpretability for MLX (Version 0.1.0) [Computer software]. https://github.com/coairesearch/mlxterp

MLA¶

Schacht, Sigurd. mlxterp: Mechanistic Interpretability for MLX. Version 0.1.0, 2025. GitHub, https://github.com/coairesearch/mlxterp.

If you use mlxterp, you may also want to cite the frameworks that inspired it:

nnsight¶

@software{nnsight2024,
  author = {Fiotto-Kaufman, Jaden and others},
  title = {nnsight: A Python package for interpreting language models},
  year = {2024},
  url = {https://nnsight.net}
}

nnterp¶

@software{nnterp2024,
  author = {Butanium},
  title = {nnterp: Unified LLM Interpretability Interface},
  year = {2024},
  url = {https://github.com/Butanium/nnterp}
}

MLX¶

@software{mlx2023,
  author = {Apple Inc.},
  title = {MLX: An array framework for Apple silicon},
  year = {2023},
  url = {https://github.com/ml-explore/mlx}
}

Acknowledgments¶

We thank:

The MLX team at Apple for creating an excellent ML framework
The nnsight and nnterp teams for pioneering clean interpretability APIs
The mechanistic interpretability community for advancing the field

Contributing¶

Contributions to mlxterp are welcome! See the Contributing Guide for guidelines.