SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability
This repository contains model checkpoints for Sparse Autoencoders (SAEs), as described in the paper SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability.
These models can be used for feature extraction.
Project page: https://saebench.xyz. For code, please see the SAEBench repository.