Householder Pseudo-Rotation: A Novel Approach to Activation Editing in LLMs with Direction-Magnitude Perspective
Abstract: Activation Editing, which involves directly editting the internal representations of large language models (LLMs) to alter their behavior and achieve desired properties, has emerged as a promising area of research. Existing works primarily treat LLMs' activations as points in space and modify them by adding steering vectors. We show that doing so would break the magnitude consistency of the activation vectors in LLMs. To overcome this shortcoming, we propose a novel editing method that views activations in terms of their directions and magnitudes. Our method, which we name \emph{Householder Pseudo-Rotation} (HPR), mimics the rotation transformation, thus preserving activation norm and resulting in an improved performance on various safety benchmarks.
Paper Type: Long
Research Area: Efficient/Low-Resource Methods for NLP
Research Area Keywords: Large Language Models, Activation Editing, Efficient Training and Inference, Rotation, Safety, Representation Engineering
Contribution Types: Model analysis & interpretability, NLP engineering experiment, Approaches to low-resource settings
Languages Studied: English
Submission Number: 2374
Loading