MoPD: Mixture-of-Prompts Distillation for Vision-Language Models