Antidote: Post-fine-tuning Safety Alignment for Large Language Models against Harmful Fine-tuning

TL;DR

Antidote is a post-fine-tuning method that prunes harmful parameters from large language models to restore safety without being affected by the hyper-parameters used during fine-tuning.

Contribution

It introduces a simple one-shot pruning approach that effectively removes harmful behaviors from LLMs regardless of fine-tuning hyper-parameters.

Findings

Reduces harmful content generation effectively.

Maintains accuracy on downstream tasks.

Robust against different fine-tuning hyper-parameters.

Abstract

Safety aligned Large Language Models (LLMs) are vulnerable to harmful fine-tuning attacks -- a few harmful data mixed in the fine-tuning dataset can break the LLMs's safety alignment. While several defenses have been proposed, our evaluation shows that existing defenses fail \textit{when some specific training hyper-parameters are chosen} -- a large learning rate or a large number of training epochs in the fine-tuning stage can easily invalidate the defense. To this end, we propose Antidote, a post-fine-tuning stage solution, which remains \textbf{\textit{agnostic to the training hyper-parameters in the fine-tuning stage}}. Antidote relies on the philosophy that by removing the harmful parameters, the harmful model can be recovered from the harmful behaviors, regardless of how those harmful parameters are formed in the fine-tuning stage. With this philosophy, we introduce a one-shot pruning stage after harmful fine-tuning to remove the harmful weights that are responsible for the generation of harmful content. Despite its embarrassing simplicity, empirical results show that Antidote can reduce harmful score while maintaining accuracy on downstream tasks. Code is available at https://github.com/git-disl/Antidote.