“perplexity filter”

baseline-defenses

Research code for evaluating defenses against adversarial attacks on aligned language models