Federated Large Language Models: Current Progress and Future Directions

TL;DR

This survey analyzes Federated Learning for Large Language Models (FedLLM), focusing on federated fine-tuning and prompt learning to address privacy, efficiency, and heterogeneity challenges. It synthesizes taxonomies, methods, and benchmarks across data and model heterogeneity, privacy threats, and communication-efficient techniques, while surveying real-world frameworks and convergence analyses. The paper highlights emerging directions, including federated pre-training, federated AI agents, multimodal FedLLMs, and LLMs tailored for FL, as well as synthetic data and capacity-augmented FL. By providing a comprehensive, up-to-date reference, it aims to guide researchers and practitioners in designing privacy-preserving, scalable FedLLMs with practical deployment considerations.

Abstract

Large language models are rapidly gaining popularity and have been widely adopted in real-world applications. While the quality of training data is essential, privacy concerns arise during data collection. Federated learning offers a solution by allowing multiple clients to collaboratively train LLMs without sharing local data. However, FL introduces new challenges, such as model convergence issues due to heterogeneous data and high communication costs. A comprehensive study is required to address these challenges and guide future research. This paper surveys Federated learning for LLMs (FedLLM), highlighting recent advances and future directions. We focus on two key aspects: fine-tuning and prompt learning in a federated setting, discussing existing work and associated research challenges. We finally propose potential directions for federated LLMs, including pre-training, federated agents, and LLMs for federated learning.