Robust Federated Finetuning of LLMs via Alternating Optimization of LoRA
This work addresses the challenge of reducing computational and communication costs in federated learning for large language models, offering a more robust fine-tuning method, though it appears incremental as it builds on existing LoRA techniques.
The paper tackles the problem of robust and efficient federated fine-tuning of large language models by proposing RoLoRA, a framework that uses alternating optimization of LoRA adapters, achieving improved performance over prior methods in experiments on models like RoBERTa-Large and Llama-2-7B across diverse tasks.
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.