CURLoRA: Stable LLM Continual Fine-Tuning and Catastrophic Forgetting Mitigation

TL;DR

CURLoRA is a novel fine-tuning method for large language models that effectively mitigates catastrophic forgetting and reduces trainable parameters by leveraging a modified CUR matrix decomposition with implicit regularization.

Contribution

It introduces a unique CUR decomposition-based approach with inverted probability selection and zero-initialized U matrix, enhancing continual learning stability and parameter efficiency.

Findings

Outperforms standard LoRA in catastrophic forgetting mitigation

Maintains model stability and performance across tasks

Reduces the number of trainable parameters significantly

Abstract

This paper introduces CURLoRA, a novel approach to fine-tuning large language models (LLMs) that leverages CUR matrix decomposition in the context of Low-Rank Adaptation (LoRA). Our method addresses two critical challenges in LLM fine-tuning: mitigating catastrophic forgetting during continual learning and reducing the number of trainable parameters. We propose a unique modification to the CUR decomposition process, utilizing inverted probabilities for column and row selection which acts as an implicit regularization, and initializing the $U$ matrix as a zero matrix, and only fine-tuning it. We demonstrate through experiments on multiple datasets that CURLoRA outperforms standard LoRA in mitigating catastrophic forgetting. It maintains model stability and performance across tasks while significantly reducing the number of trainable parameters. Our results show that CURLoRA achieves very good and stable task accuracy while maintaining base model's perplexity scores fixed compared to LoRA upon continual fine-tuning, particularly in scenarios with limited data.