List your company

Adversarial Toxicity Refusal on LLaMA-2 Chatbot Offensive category

47.7RTRAdv. attack
-1.711.12523.9536.775Jul 8, 2025

Evaluation results

MethodLinks
Adv. attackDefense setting=Adv. attack2025.0747.75.299.8-2.86
No attackDefense setting=No attack2025.078.84.2710.2—
Optimus (NH)Defense setting=Optimus (NH)2025.070.26.1710.9—