From Backward Spreading to Forward Replay: Revisiting Target Construction in LLM Parameter Editing

TL;DR

This paper critically examines the foundations of backward spreading in LLM parameter editing and introduces a forward propagation approach that improves target accuracy and compatibility across layers.

Contribution

It systematically investigates backward spreading and proposes a forward propagation method for more accurate and compatible layer-wise target hidden-states in LLM editing.

Findings

Forward propagation achieves more accurate layer-wise targets.

The new method maintains the same computational complexity as existing approaches.

It is simple to integrate without disrupting existing editing pipelines.

Abstract

LLM parameter editing methods commonly rely on computing an ideal target hidden-state at a target layer (referred as anchor point) and distributing the target vector to multiple preceding layers (commonly known as backward spreading) for cooperative editing. Although widely used for a long time, its underlying basis have not been systematically investigated. In this paper, we first conduct a systematic study of its foundations, which helps clarify its capability boundaries, practical considerations, and potential failure modes. Then, we propose a simple and elegant alternative that replaces backward spreading with forward-propagation. Instead of optimizing the target at the last editing layer, we optimize the anchor point at the first editing layer, and then propagate it forward to obtain accurate and mutually compatible target hidden-states for all subsequent editing layers. This approach achieves the same computational complexity as existing methods while producing more accurate layer-wise targets. Our method is simple, without interfering with either the computation of the initial target hidden state or any other components of the subsequent editing pipeline, and thus constituting a benefit for a wide range of LLM parameter editing methods.