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

Muhammad Fawi

返回文献探索

Paper · arXiv 2408.14572

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

Muhammad Fawi

8 upvotesAugust 26, 2024arXiv 预印本

AI 摘要

CURLoRA, a method that uses CUR matrix decomposition in Low-Rank Adaptation (LoRA), effectively mitigates catastrophic forgetting and reduces trainable parameters in fine-tuning large language models.

CUR matrix decompositionLow-Rank Adaptation (LoRA)catastrophic forgettingcontinual learning

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.