CNT: Safety-oriented Function Reuse across LLMs via Cross-Model Neuron Transfer

TL;DR

This paper introduces Cross-Model Neuron Transfer (CNT), a post-hoc method for reusing safety-related functionalities across large language models by transferring minimal neurons, improving safety adaptation with minimal performance loss.

Contribution

CNT enables modular safety functionality transfer between LLMs at the neuron level, supporting both addition and deletion of safety features in a post-hoc manner.

Findings

Achieves safety functionality transfer with less than 1% performance degradation.

Outperforms five baseline methods across multiple safety tasks.

Demonstrates generality and effectiveness in diverse LLMs.

Abstract

The widespread deployment of large language models (LLMs) calls for post-hoc methods that can flexibly adapt models to evolving safety requirements. Meanwhile, the rapidly expanding open-source LLM ecosystem has produced a diverse collection of models that already exhibit various safety-related functionalities. This motivates a shift from constructing safety functionality from scratch to reusing existing functionality from external models, thereby avoiding costly data collection and training procedures. In this paper, we present Cross-Model Neuron Transfer (CNT), a post-hoc method that reuses safety-oriented functionality by transferring a minimal subset of neurons from an open-source donor LLM to a target LLM. By operating at the neuron level, CNT enables modular function-level adaptation, supporting both function addition andfunction deletion. We evaluate CNT on seven popular LLMs across three representative applications: safety disalignment, alignment enhancement, and bias removal. Experimental results show that CNT achieves targeted safety-oriented functionality transfer with minimal performance degradation (less than 1% for most models), consistently outperforming five baselines, demonstrating its generality and practical effectiveness.