Anchor-based Robust Finetuning of Vision-Language Models | Read Paper on Bytez