Saev

saev is a package for training sparse autoencoders (SAEs) on vision transformers (ViTs) in PyTorch. It also includes an interactive webapp for looking through a trained SAE's features.

API reference docs are available below, as well as the source code on GitHub.

My logbook is a set of notes that might also be useful.

I am working on some web apps so anyone can play with SAEs for vision without needing to run any code themselves. The first of these web apps is here. Please don't share it publicly yet.

Package Docs

saev

Package to train SAEs for vision models.

contrib

Individual sub-packages not related to the core package.