SECURA: Sigmoid-Enhanced CUR Decomposition with Uninterrupted Retention and Low-Rank Adaptation in Large Language Models
This work addresses the challenge of efficient and stable fine-tuning for large language models, offering a solution that mitigates forgetting and improves performance, though it appears incremental as an enhancement to existing LoRA methods.
The paper tackles the problem of catastrophic forgetting and computational inefficiency in fine-tuning large language models by proposing SECURA, a novel PEFT variant that achieves average improvements of 3.59% on MCQ tasks and 2.51% on QA tasks, while maintaining over 70% accuracy in knowledge retention tests.
With the rapid development of large language models (LLMs), fully fine-tuning (FT) these models is becoming increasingly infeasible due to high computational demands. Moreover, FT also increases the risk of catastrophic forgetting. As an alternative, Low-Rank Adaptation (LoRA) has been proposed. By fine-tuning only a small subset of parameters, LoRA achieves performance similar to FT while significantly reducing resource requirements. However, since LoRA inherits FT's design, the issue of catastrophic forgetting still remains. To address these limitations, we propose SECURA: Sigmoid-Enhanced CUR Decomposition LoRA, a novel PEFT variant designed to mitigate catastrophic forgetting while improving fine-tuning performance. Our method introduces a novel normalization technique, Sigmoid-based Magnitude Norm (S-MagNorm), which enhances parameter retention and fine-tuning efficiency. SECURA has been evaluated on a diverse range of tasks, including mathematical problem-solving (GSM8K), complex question-answering (CNNDM), translation (NewsDE), and complex multiple-choice reasoning (LogiQA). Experimental results demonstrate that it achieves an average fine-tuning improvement of 3.59% across four MCQ tasks and 2.51% across five QA tasks on Gemma2 2B, Qwen2 1.5B, Qwen2 7B, Llama3 8B, and Llama3.1 8B, outperforming DoRA. Additionally, SECURA demonstrates superior knowledge retention capabilities, achieving state-of-the-art performance in 16 continual learning tests and maintaining more than 70% accuracy on LLMs' basic knowledge compared to Experience Replay (ER), sequential learning (SEQ), EWC, I-LoRA, and CUR-LoRA.