Differentially Private Low-Rank Adaptation of Large Language Model Using Federated Learning