Release interpretability datasets on Hugging Face

[NielsRogge]

Hi @belindal 🤗

Niels here from the open-source team at Hugging Face. I discovered your work through Hugging Face's daily papers as yours got featured: https://huggingface.co/papers/2511.08579.
The paper page lets people discuss about your paper and lets them find artifacts about it (your models and datasets for instance), you can also claim
the paper as yours which will show up on your public profile at HF, add Github and project page URLs.

It is fantastic to see that you have already hosted most of the explainer models and the intervention results on the Hub under the Transluce organization! It'd be great to also make the underlying feature and activation datasets (currently on S3) available on the 🤗 hub to improve their discoverability and visibility.

Uploading datasets

Moving the datasets from AWS S3 to 🤗 would allow people to do:

from datasets import load_dataset

# Example for the FineWeb activations
dataset = load_dataset("Transluce/fineweb-llama-activations")

Besides that, there's the dataset viewer which allows people to quickly explore the data structure in the browser. Since these datasets involve activations and natural language explanations, the viewer could be a great way for the community to browse through the "privileged access" signals you've extracted.

See here for a guide on uploading: https://huggingface.co/docs/datasets/loading.

Let me know if you're interested or need any help regarding this!

Cheers,

Niels
ML Engineer @ HF 🤗