Route Sparse Autoencoder to Interpret Large Language Models
This work addresses the challenge of multi-layer feature extraction for mechanistic interpretability in LLMs, offering a scalable method for researchers and practitioners, though it is incremental as it builds on existing sparse autoencoder techniques.
The paper tackles the problem of extracting interpretable features from multiple layers in large language models, introducing Route Sparse Autoencoder (RouteSAE) which, under a sparsity constraint of 64, extracts 22.5% more features and achieves a 22.3% higher interpretability score than baseline methods.
Mechanistic interpretability of large language models (LLMs) aims to uncover the internal processes of information propagation and reasoning. Sparse autoencoders (SAEs) have demonstrated promise in this domain by extracting interpretable and monosemantic features. However, prior works primarily focus on feature extraction from a single layer, failing to effectively capture activations that span multiple layers. In this paper, we introduce Route Sparse Autoencoder (RouteSAE), a new framework that integrates a routing mechanism with a shared SAE to efficiently extract features from multiple layers. It dynamically assigns weights to activations from different layers, incurring minimal parameter overhead while achieving high interpretability and flexibility for targeted feature manipulation. We evaluate RouteSAE through extensive experiments on Llama-3.2-1B-Instruct. Specifically, under the same sparsity constraint of 64, RouteSAE extracts 22.5% more features than baseline SAEs while achieving a 22.3% higher interpretability score. These results underscore the potential of RouteSAE as a scalable and effective method for LLM interpretability, with applications in feature discovery and model intervention. Our codes are available at https://github.com/swei2001/RouteSAEs.