Training-free Lexical Backdoor Attacks on Language Models

Open in new window