FedFit: Federated Fine-Tuning of LLMs via Vector-Bank Parameterization and Quantization

2026-10-01

AI Summary

This work addresses the substantial communication overhead and LoRA aggregation conflicts inherent in federated fine-tuning of large language models by proposing a parameterized reconstruction method based on a decoupled shared vector library. Through an alternating optimization strategy and a residual spectral aggregation mechanism, the approach effectively resolves Sum-of-Products (SoP) versus Product-of-Sums (PoS) conflicts. Furthermore, it integrates block quantization with error feedback to compress transmitted data, thereby enabling efficient and privacy-preserving fine-tuning. Experiments conducted on Qwen2.5 demonstrate that the proposed method achieves over a hundredfold communication compression ratio while yielding perplexity performance comparable to standard centralized training. Additionally, the framework is supported by rigorous convergence guarantees.

Abstract

Federated Learning (FL) enables privacy-preserving fine-tuning of Large Language Models (LLMs), yet the massive communication overhead remains a critical bottleneck. Furthermore, applying Low-Rank Adaptation (LoRA) in FL faces a fundamental "aggregation dilemma" between the accurate Sum-of-Products (SoP) and the communication-efficient Product-of-Sums (PoS) implementations. To tackle these challenges, we propose FedFit. First, to significantly reduce communication overhead, we introduce a disjoint shared vector-bank parameterization that reconstructs high-dimensional adapter matrices from two compact and disjoint global vector banks. Second, to address the aggregation dilemma, we devise an alternating optimization schedule. By cycling between decoupled single-bank updates (which allow for accurate aggregation) and joint updates corrected by a Residual Spectral Aggregation mechanism, we resolve the conflict between SoP and PoS. Additionally, we integrate blockwise quantization with client-side error feedback to further compress the transmitted vectors. Furthermore, we establish theoretical convergence guarantees for the proposed algorithm. Extensive experiments on Qwen2.5 models demonstrate that FedFit achieves perplexity performance comparable to standard federated LoRA methods, while providing compression ratios up to 100x higher.

Problem

• Federated Learning
• Large Language Models
• Fine-Tuning
• Communication Overhead
• Low-Rank Adaptation

Innovation

• Federated Learning
• Low-Rank Adaptation
• Vector-Bank Parameterization
• Alternating Optimization
• Quantization