Release code and artifacts for "Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models" on Hugging Face

[NielsRogge]

Hi @mateuszpach 🤗

I'm Niels from the Hugging Face open-source team. I came across your interesting work on Sparse Autoencoders for Vision-Language Models through our daily paper feature: https://huggingface.co/papers/2504.02821. Claiming your paper on the page allows you to add links to your GitHub repo, project page, and models, boosting visibility and facilitating discussion.

Your README mentions that scripts to run the code are "To be added soon." Once the code and checkpoints are ready, it would be fantastic to host them on the Hugging Face Hub. This significantly improves discoverability and allows seamless integration with the Hugging Face ecosystem.

Here's how you can upload your models: https://huggingface.co/docs/hub/models-uploading. For custom PyTorch models, the PyTorchModelHubMixin (docs) adds convenient from_pretrained and push_to_hub methods. If you prefer, direct uploads or hf_hub_download (docs) are also supported. We recommend separate repositories for each checkpoint to track download statistics effectively. Once uploaded, we can link them directly to your paper page.

Let me know if you have any questions or need assistance with the process. We're happy to help!

Cheers,

Niels
ML Engineer @ HF 🤗

[ADITHYAG73]

This is an amazing work. Looking forward to it on hugging face.