Enregistré dans:
Détails bibliographiques
Auteurs principaux: Wang, Chenxu, Lyu, Yilin, Sun, Zicheng, Jing, Liping
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2507.02503
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866916828823945216
author Wang, Chenxu
Lyu, Yilin
Sun, Zicheng
Jing, Liping
author_facet Wang, Chenxu
Lyu, Yilin
Sun, Zicheng
Jing, Liping
contents Continual fine-tuning of Large Language Models (LLMs) is hampered by the trade-off between efficiency and expressiveness. Low-Rank Adaptation (LoRA) offers efficiency but constrains the model's ability to learn new tasks and transfer knowledge due to its low-rank nature and reliance on explicit parameter constraints. We propose GORP (Gradient LOw Rank Projection) for Continual Learning, a novel training strategy that overcomes these limitations by synergistically combining full and low-rank parameters and jointly updating within a unified low-rank gradient subspace. GORP expands the optimization space while preserving efficiency and mitigating catastrophic forgetting. Extensive experiments on continual learning benchmarks demonstrate GORP's superior performance compared to existing state-of-the-art approaches. Code is available at https://github.com/Wcxwcxw/GORP.
format Preprint
id arxiv_https___arxiv_org_abs_2507_02503
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Continual Gradient Low-Rank Projection Fine-Tuning for LLMs
Wang, Chenxu
Lyu, Yilin
Sun, Zicheng
Jing, Liping
Machine Learning
Artificial Intelligence
Computational Engineering, Finance, and Science
Continual fine-tuning of Large Language Models (LLMs) is hampered by the trade-off between efficiency and expressiveness. Low-Rank Adaptation (LoRA) offers efficiency but constrains the model's ability to learn new tasks and transfer knowledge due to its low-rank nature and reliance on explicit parameter constraints. We propose GORP (Gradient LOw Rank Projection) for Continual Learning, a novel training strategy that overcomes these limitations by synergistically combining full and low-rank parameters and jointly updating within a unified low-rank gradient subspace. GORP expands the optimization space while preserving efficiency and mitigating catastrophic forgetting. Extensive experiments on continual learning benchmarks demonstrate GORP's superior performance compared to existing state-of-the-art approaches. Code is available at https://github.com/Wcxwcxw/GORP.
title Continual Gradient Low-Rank Projection Fine-Tuning for LLMs
topic Machine Learning
Artificial Intelligence
Computational Engineering, Finance, and Science
url https://arxiv.org/abs/2507.02503