Towards Federated Low-Rank Adaptation of Language Models with Rank Heterogeneity

TL;DR

This paper introduces a replication-based padding strategy for federated LoRA that addresses rank heterogeneity issues, improving convergence speed and model performance in federated language model fine-tuning.

Contribution

It proposes a novel replication-based padding method to handle heterogeneous ranks in federated LoRA, enhancing stability and effectiveness.

Findings

Replication-based padding improves convergence speed.

The method enhances global model accuracy.

It effectively manages rank heterogeneity issues.

Abstract

Low-rank adaptation (LoRA) offers an efficient alternative to full-weight adaptation in federated fine-tuning of language models, significantly reducing computational costs. By adjusting ranks for each client, federated LoRA enables flexible resource allocation. However, we observe that heterogeneous ranks among clients lead to unstable performance. Our analysis attributes this instability to the conventional zero-padding aggregation strategy, which dilutes information from high-rank clients during model aggregation. To address this issue, we propose a replication-based padding strategy that better retains valuable information from clients with high-quality data. Empirically, this approach accelerates convergence and enhances the global model's predictive performance.