Benchmarking Robustness of Adaptation Methods on Pre-trained Vision-Language Models Shuo Chen 1,3 Jindong Gu2 Zhen Han 1 Y unpu Ma1,3