Fetching the paper…

Defending Pre-trained Language Models from Adversarial Word Substitutions Without Performance Sacrifice · Around