SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability
This provides a standardized framework for researchers to systematically compare SAE architectures and training methods, addressing a bottleneck in interpretability research.
The authors tackled the lack of reliable evaluation for sparse autoencoders (SAEs) in language model interpretability by introducing SAEBench, a comprehensive benchmark with eight diverse metrics, revealing that gains on proxy metrics do not reliably translate to better practical performance, such as Matryoshka SAEs substantially outperforming others on feature disentanglement metrics with scaling advantages.
Sparse autoencoders (SAEs) are a popular technique for interpreting language model activations, and there is extensive recent work on improving SAE effectiveness. However, most prior work evaluates progress using unsupervised proxy metrics with unclear practical relevance. We introduce SAEBench, a comprehensive evaluation suite that measures SAE performance across eight diverse metrics, spanning interpretability, feature disentanglement and practical applications like unlearning. To enable systematic comparison, we open-source a suite of over 200 SAEs across eight recently proposed SAE architectures and training algorithms. Our evaluation reveals that gains on proxy metrics do not reliably translate to better practical performance. For instance, while Matryoshka SAEs slightly underperform on existing proxy metrics, they substantially outperform other architectures on feature disentanglement metrics; moreover, this advantage grows with SAE scale. By providing a standardized framework for measuring progress in SAE development, SAEBench enables researchers to study scaling trends and make nuanced comparisons between different SAE architectures and training methodologies. Our interactive interface enables researchers to flexibly visualize relationships between metrics across hundreds of open-source SAEs at: www.neuronpedia.org/sae-bench