PMET: Precise Model Editing in a Transformer

TL;DR

PMET introduces a novel approach to model editing in transformers by separately optimizing MHSA and FFN components, leading to more precise updates and state-of-the-art results on knowledge editing benchmarks.

Contribution

The paper presents PMET, a new method that optimizes transformer hidden states for more accurate model editing by leveraging the distinct roles of MHSA and FFN.

Findings

MHSA encodes general knowledge extraction patterns.

Optimizing only FFN hidden states improves editing precision.

PMET achieves state-of-the-art results on COUNTERFACT and zsRE datasets.

Abstract

Model editing techniques modify a minor proportion of knowledge in Large Language Models (LLMs) at a relatively low cost, which have demonstrated notable success. Existing methods assume Transformer Layer (TL) hidden states are values of key-value memories of the Feed-Forward Network (FFN). They usually optimize the TL hidden states to memorize target knowledge and use it to update the weights of the FFN in LLMs. However, the information flow of TL hidden states comes from three parts: Multi-Head Self-Attention (MHSA), FFN, and residual connections. Existing methods neglect the fact that the TL hidden states contains information not specifically required for FFN. Consequently, the performance of model editing decreases. To achieve more precise model editing, we analyze hidden states of MHSA and FFN, finding that MHSA encodes certain general knowledge extraction patterns. This implies that MHSA weights do not require updating when new knowledge is introduced. Based on above findings, we introduce PMET, which simultaneously optimizes Transformer Component (TC, namely MHSA and FFN) hidden states, while only using the optimized TC hidden states of FFN to precisely update FFN weights. Our experiments demonstrate that PMET exhibits state-of-the-art performance on both the COUNTERFACT and zsRE datasets. Our ablation experiments substantiate the effectiveness of our enhancements, further reinforcing the finding that the MHSA encodes certain general knowledge extraction patterns and indicating its storage of a small amount of factual knowledge. Our code is available at https://github.com/xpq-tech/PMET.