Dissecting Fine-Tuning Unlearning in Large Language Models

TL;DR

This paper critically examines fine-tuning-based unlearning in large language models, revealing that such methods do not truly erase knowledge but instead alter retrieval processes, impacting overall model behavior.

Contribution

It uncovers the limitations of current unlearning techniques, highlighting the role of MLP coefficients and showing their effects on model behavior and knowledge retention.

Findings

Unlearning methods do not genuinely erase embedded knowledge.

MLP coefficients are key to controlling model behavior.

Unlearning impacts unrelated knowledge and capabilities.

Abstract

Fine-tuning-based unlearning methods prevail for preventing targeted harmful, sensitive, or copyrighted information within large language models while preserving overall capabilities. However, the true effectiveness of these methods is unclear. In this work, we delve into the limitations of fine-tuning-based unlearning through activation patching and parameter restoration experiments. Our findings reveal that these methods alter the model's knowledge retrieval process, providing further evidence that they do not genuinely erase the problematic knowledge embedded in the model parameters. Instead, the coefficients generated by the MLP components in the model's final layer are the primary contributors to these seemingly positive unlearning effects, playing a crucial role in controlling the model's behaviors. Furthermore, behavioral tests demonstrate that this unlearning mechanism inevitably impacts the global behavior of the models, affecting unrelated knowledge or capabilities. The code is released at https://github.com/yihuaihong/Dissecting-FT-Unlearning.