Vision Transformer Adapters for Generalizable Multitask Learning