AtP*: An efficient and scalable method for localizing LLM behaviour to
  components

János Kramár; Tom Lieberum; Rohin Shah; Neel Nanda

返回文献探索

Paper · arXiv 2403.00745

AtP*: An efficient and scalable method for localizing LLM behaviour to components

János Kramár, Tom Lieberum, Rohin Shah, Neel Nanda

13 upvotesMarch 1, 2024arXiv 预印本

AI 摘要

Attribution Patching (AtP) is a scalable method for approximating causal attributions in large language models, outperforming other techniques including its improved variant AtP*, which also provides a method to bound false negatives.

Activation PatchingAttribution PatchingAtPAtP*Large Language Models (LLMs)causal attributionsgradient-based approximationfalse negatives

Abstract

Activation Patching is a method of directly computing causal attributions of behavior to model components. However, applying it exhaustively requires a sweep with cost scaling linearly in the number of model components, which can be prohibitively expensive for SoTA Large Language Models (LLMs). We investigate Attribution Patching (AtP), a fast gradient-based approximation to Activation Patching and find two classes of failure modes of AtP which lead to significant false negatives. We propose a variant of AtP called AtP*, with two changes to address these failure modes while retaining scalability. We present the first systematic study of AtP and alternative methods for faster activation patching and show that AtP significantly outperforms all other investigated methods, with AtP* providing further significant improvement. Finally, we provide a method to bound the probability of remaining false negatives of AtP* estimates.