Separate The Wheat From The Chaff: Model Deficiency Unlearning Via Parameter-efficient Module Operation
Hu Xinshuo, Li Dongfang, Hu Baotian, Zheng Zihao, Liu Zhenyu, Zhang Min. Arxiv 2023
[Paper]
Large language models (LLMs) have been widely used in various applications
but are known to suffer from issues related to untruthfulness and toxicity.
While parameter-efficient modules (PEMs) have demonstrated their effectiveness
in equipping models with new skills, leveraging PEMs for deficiency unlearning
remains underexplored. In this work, we propose a PEMs operation approach,
namely Extraction-before-Subtraction (Ext-Sub), to enhance the truthfulness and
detoxification of LLMs through the integration of expert'' PEM and
anti-expert’’ PEM. Remarkably, even anti-expert PEM possess valuable
capabilities due to their proficiency in generating fabricated content, which
necessitates language modeling and logical narrative competence. Rather than
merely negating the parameters, our approach involves extracting and
eliminating solely the deficiency capability within anti-expert PEM while
preserving the general capabilities. To evaluate the effectiveness of our
approach in terms of truthfulness and detoxification, we conduct extensive
experiments on LLMs, encompassing additional abilities such as language
modeling and mathematical reasoning. Our empirical results demonstrate that our
approach effectively improves truthfulness and detoxification, while largely
preserving the fundamental abilities of LLMs.
Similar Work