Low-Rank Adaptation: Exploring LORA for Efficient Language Model Fine-Tuning