Crafting Reversible SFT Behaviors in Large Language Models

TL;DR

This paper introduces methods to create sparse, causally necessary subnetworks within large language models that encode specific behaviors, enabling deliberate control and reversal of these behaviors at inference time.

Contribution

It proposes Loss-Constrained Dual Descent and SFT-Eraser techniques to embed and reverse behaviors via sparse carriers, advancing behavior control in language models.

Findings

Sparse carriers preserve target behaviors and enable reversal.

Structure rather than trigger design is key for behavior reversal.

Carriers are causally necessary for SFT-induced behaviors.

Abstract

Supervised fine-tuning (SFT) induces new behaviors in large language models, yet imposes no structural constraint on how these behaviors are distributed within the model. Existing behavior interpretation methods, such as circuit attribution approaches, identify sparse subnetworks correlated with SFT-induced behaviors post-hoc. However, such correlations do not imply *causal necessity*, limiting the ability to selectively control SFT-induced behaviors at inference time. We pursue an alternative by asking: can an SFT-induced behavior be deliberately compressed into a sparse, mechanistically necessary subnetwork, termed a *carrier*, while remaining controllable at inference time without weight modification? We propose (a) **Loss-Constrained Dual Descent (LCDD)**, which constructs such carriers by jointly optimizing routing masks and model weights under an explicit utility budget, and (b) **SFT-Eraser**, a soft prompt optimized via activation matching on extracted carrier channels, to reverse the SFT-induced behavior. Across safety, fixed-response, and style behaviors on multiple model families, LCDD yields sparse carriers that preserve target behaviors while enabling strong reversion when triggered by SFT-Eraser. Ablations further establish that the sparse structure is the key precondition for reversal: the same trigger optimization fails on standard SFT models, confirming that structure rather than trigger design is the operative factor. These results provide direct evidence that the learned carriers are causally necessary for the behaviors, pointing to a new direction for systematically localizing and selectively suppressing SFT-induced behaviors in deployed models.