Robust Federated Finetuning of LLMs via Alternating Optimization of LoRA

TL;DR

This paper introduces RoLoRA, a federated fine-tuning framework for large language models that employs alternating optimization of LoRA adapters, improving expressiveness and robustness through theoretical insights and extensive experiments.

Contribution

It proposes a novel federated fine-tuning method, RoLoRA, that optimizes both up and down projection matrices in LoRA, backed by theoretical analysis and comprehensive experimental validation.

Findings

RoLoRA outperforms prior methods in federated settings.

Learning both projection matrices enhances model expressiveness.

Theoretical analysis supports the importance of dual matrix learning.

Abstract

Parameter-Efficient Fine-Tuning (PEFT) methods like Low-Rank Adaptation (LoRA) optimize federated training by reducing computational and communication costs. We propose RoLoRA, a federated framework using alternating optimization to fine-tune LoRA adapters. Our approach emphasizes the importance of learning up and down projection matrices to enhance expressiveness and robustness. We use both theoretical analysis and extensive experiments to demonstrate the advantages of RoLoRA over prior approaches that either generate imperfect model updates or limit expressiveness of the model. We provide a theoretical analysis on a linear model to highlight the importance of learning both the down-projection and up-projection matrices in LoRA. We validate the insights on a non-linear model and separately provide a convergence proof under general conditions. To bridge theory and practice, we conducted extensive experimental evaluations on language models including RoBERTa-Large, Llama-2-7B on diverse tasks and FL settings to demonstrate the advantages of RoLoRA over other methods.