MMA: Multi-Modal Adapter for Vision-Language Models | Read Paper on Bytez