Mitigating Catastrophic Forgetting in Large Language Models with Forgetting-aware Pruning
Keywords: Large Language Models, Catastrophic Forgetting, Neural Network Pruning
TL;DR: We propose Forgetting-Aware Pruning Metric, a novel pruning-based approach to balance Catastrophic Forgetting and downstream task performance for LLMs.
Abstract: Recent advancements in Large Language Models (LLMs) have demonstrated remarkable capabilities across a wide range of tasks. These models are typically pretrained on extensive corpora and subsequently fine-tuned on task-specific datasets. However, during the fine-tuning process, LLMs often suffer from Catastrophic Forgetting (CF), wherein previously acquired general knowledge is lost. Traditional approaches to mitigating CF often rely on data replay, which may not be viable when the original training data is inaccessible. Additionally, methods that alter the training process or the model architecture can increase complexity and detract from the accuracy of downstream tasks, thus limiting their generalizability. In this paper, we propose Forgetting-Aware Pruning Metric (FAPM), a novel pruning-based approach to balance CF and downstream task performance. Our investigation reveals that the degree to which task vectors (i.e., the subtraction of pre-trained weights from the weights fine-tuned on downstream tasks) overlap with pre-trained model parameters is a critical factor for CF. Motivated by this insight, FAPM employs the ratio of the task vector to pre-trained model parameters as a metric to quantify CF, integrating this measure into the pruning criteria. Importantly, FAPM does not necessitate modifications to the training process or model architecture, nor does it require any auxiliary data. We conducted extensive experiments across six datasets encompassing natural language inference, question answering, reading comprehension, and cloze tests. The results demonstrate that FAPM limits CF to just 1% while maintaining 99% accuracy on downstream tasks, rendering FAPM highly competitive relative to the state-of-the-art methods that involve modifications to the training process.
Primary Area: transfer learning, meta learning, and lifelong learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 257
Loading