This is the prototypical implementation for the paper "black-box adversarial attack guided by model behavior for programming pre-trained language models".
For simplicity, we only provide the prototypical implementation of the codebert+authority task, the prototypical implementation of other tasks is basically the same.