arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2602.16936cs.DC
收藏

异构联邦微调与并行一秩适应

Heterogeneous Federated Fine-Tuning with Parallel One-Rank Adaptation

University of Nevada, Reno(内华达大学雷诺分校)

Zikai Zhang, Rui Hu, Jiahao Xu

AI总结:

Fed-PLoRA通过并行一秩适应和Select-N-Fold策略,解决异构联邦微调中的初始化和聚合噪声问题,提升LLM微调的准确性和效率。

摘要:

大型语言模型(LLMs)通过微调在适应下游任务方面表现出显著的有效性。联邦学习(FL)通过使用低秩适应(LoRA)在分布式客户端之间实现协作微调,同时通过避免原始数据共享来保护数据隐私。然而,当客户端具有异构资源并因此采用不同的LoRA秩时,实际部署会面临挑战,这导致初始化和聚合噪声显著,从而影响性能。为了解决这些挑战,我们提出了Fed-PLoRA,一种新颖的轻量级异构联邦微调(FFT)框架。Fed-PLoRA引入了并行一秩适应(PLoRA),一种新的LoRA变体,它用多个并行一秩模块替代经典的多秩LoRA模块,并且引入了一种新的Select-N-Fold策略,在本地训练之前将未训练的PLoRA模块折叠到预训练权重中,从而适应异构客户端资源。我们提供了Fed-PLoRA初始化和聚合噪声的统一分析,并展示了其如何克服现有最先进方法的局限性。在多样化的LLM微调任务上进行了广泛的实验,结果表明Fed-PLoRA在准确性和效率上都一致优于现有方法。代码可在https://github.com/TNI-playground/Fed-PLoRA上获得。

英文摘要:

Large Language Models (LLMs) have demonstrated remarkable effectiveness in adapting to downstream tasks through fine-tuning. Federated Learning (FL) extends this capability by enabling collaborative fine-tuning across distributed clients using Low-Rank Adaptation (LoRA), while preserving data privacy by avoiding raw data sharing. However, practical deployments face challenges when clients have heterogeneous resources and thus adopt different LoRA ranks, leading to substantial initialization and aggregation noise that undermines performance. To address these challenges, we propose Fed-PLoRA, a novel lightweight heterogeneous federated fine-tuning (FFT) framework. Fed-PLoRA introduces Parallel One-Rank Adaptation (PLoRA), a new LoRA variant that replaces the classic multi-rank LoRA module with multiple parallel one-rank modules, and a novel Select-N-Fold strategy that folds untrained PLoRA modules into the pre-trained weights before local training, thereby accommodating heterogeneous client resources. We provide a unified analysis of initialization and aggregation noise of Fed-PLoRA and demonstrate how it addresses the limitations of state-of-the-art methods. Extensive experiments on diverse LLM fine-tuning tasks demonstrate that Fed-PLoRA consistently outperforms existing methods in both accuracy and efficiency. The code is available at https://github.com/TNI-playground/Fed-PLoRA.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑