Top-K SAE

Creator

Creator

Seonglae Cho

Created

Created

2024 Nov 18 20:25

Editor

Editor

Seonglae Cho

Edited

Edited

2025 Jun 16 22:3

Refs

Refs

2024,
Leo Gao et al.

Limit each input token to have a maximum of k feature activations

Problem: Some tokens can be easily reconstructed, but others need more features

OpenAI Top-k SAE

https://cdn.openai.com/papers/sparse-autoencoders.pdf

Feature Browser

Web site created using create-react-app

https://openaipublic.blob.core.windows.net/sparse-autoencoder/sae-viewer/index.html

Recommendations

////////////