The official code for "Adversarially Pretrained Transformers may be Universally Robust In-Context Learners" [S. Kumano et al., preprint]. - View it on GitHub