rank-heterogeneity

FedRoRA is a novel framework for personalized federated LoRA fine-tuning that addresses rank heterogeneity and data heterogeneity in federated learning by decoupling adaptation into shared global directions and personalized magnitudes.