Beyond Activation Patterns: A Weight-Based Out-of-Context Explanation of Sparse Autoencoder Features

TL;DR

This paper introduces a weight-based interpretation method for sparse autoencoders that reveals how features function in language models without relying on activation data, uncovering their roles in token prediction and attention mechanisms.

Contribution

It presents a novel weight-based framework for interpreting SAE features, addressing the limitations of activation-based methods and providing new insights into feature roles in language models.

Findings

1/4 of features directly predict output tokens

Features participate in attention mechanisms with depth-dependent structure

Semantic and non-semantic features have distinct attention distribution profiles

Abstract

Sparse autoencoders (SAEs) have emerged as a powerful technique for decomposing language model representations into interpretable features. Current interpretation methods infer feature semantics from activation patterns, but overlook that features are trained to reconstruct activations that serve computational roles in the forward pass. We introduce a novel weight-based interpretation framework that measures functional effects through direct weight interactions, requiring no activation data. Through three experiments on Gemma-2 and Llama-3.1 models, we demonstrate that (1) 1/4 of features directly predict output tokens, (2) features actively participate in attention mechanisms with depth-dependent structure, and (3) semantic and non-semantic feature populations exhibit distinct distribution profiles in attention circuits. Our analysis provides the missing out-of-context half of SAE feature interpretability.