Cross-Modal Adapter: Parameter-Efficient Transfer Learning Approach for Vision-Language Models