SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability

Devs

SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability | Read Paper on Bytez