Year: 2024
Type: preprint
Abstract: Understanding and shaping the behaviour of Large Language Models (LLMs) is increasingly important as applications become more powerful and more frequently adopted. This paper introduces a machine unlearning method specifically designed for LLMs. We introduce a selective pruning method for LLMs that removes neurons based on their relative importance on a targeted capability compared to overall network performance. This approach is a compute- and data-efficient method for identifying and removing neurons that enable specific behaviours. Our findings reveal that both feed-forward and attention neurons in LLMs are specialized; that is, for specific tasks, certain neurons are more crucial than others. Code from all experiments is available at https://github.com/nickypro/selective-pruning
Source: arXiv (Cornell University)
Authors: Pochinkov, Nicholas, Schoots, Nandi
Institution: King's College London
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4392487615