abhishek-040010/sae-for-vlms

Decoding Bias in Vision-Language Models using Sparse Autoencoders

This repository contains the trained weights for top-k and batch top-k sparse autoencoders trained for my thesis on the Imagenet Training split. The models are implemented from https://github.com/saprmarks/dictionary_learning

Code: https://github.com/abhishek-agrl/sae-for-vlms