List your company

Targeted Refusal on Llama2-7B Generation Evaluation Set

93.14Completion Accuracy (CA)Vanilla
79.942483.368786.79590.2213Nov 16, 2025

Evaluation results

MethodLinks
VanillaAttack=CBA2025.1193.1489.36
VanillaAttack=VPI2025.1191.2597.36
VanillaAttack=Sleeper Agent2025.1190.8896.64
OURSAttack=VPI2025.1190.114.45
GracCeFulAttack=CBA2025.1189.3720.95
CleanGenAttack=CBA2025.1188.9224.87
OURSAttack=CBA2025.1188.4915.4
OURSAttack=Sleeper Agent2025.1188.1411.37
CleanGenAttack=VPI2025.1187.2928.04
GracCeFulAttack=VPI2025.1187.1520.17
GracCeFulAttack=Sleeper Agent2025.1186.7821.45
CleanGenAttack=Sleeper Agent2025.1186.3425.8
MuScleLoRAAttack=VPI2025.1183.1519.57
MuScleLoRAAttack=CBA2025.1182.3719.2
MuScleLoRAAttack=Sleeper Agent2025.1180.4516.44