This is the official code base for our NeurIPS 2021 paper (Dataset and benchmark track, Oral presentation, 3.3% accepted rate) Adversarial GLUE: A Multi-Task Benchmark for Robustness Evaluation of ...