List your company

Targeted Refusal on Targeted Refusal LLaMA2-7B-Chat (test)

100BadNetsFT
3.841628.805853.7778.7342Mar 13, 2026

Evaluation results

MethodLinks
FTThreat Model=Full Params2026.0310076.2868.4694.4274.1582.66
No DefenseThreat Model=N/A2026.0398.9473.9963.3195.8377.9882.01
PruningThreat Model=Lora Adapter2026.0394.574.7862.4593.3373.8679.78
Fine-PruningThreat Model=Lora Adapter2026.0390.6752.6648.3482.3153.0465.36
PruningThreat Model=Full Params2026.0384.6839.5255.5886.8862.3765.81
QuantizationThreat Model=Full Params2026.0368.3232.8418.2964.0234.3343.56
Fine-PruningThreat Model=Full Params2026.0359.0927.6236.8456.3348.3245.64
FTThreat Model=Lora Adapter2026.0325.1844.5642.3884.3765.2352.34
CROWThreat Model=Full Params2026.0321.9343.3340.5388.6662.5751.4
BD-VAXThreat Model=Lora Adapter2026.0310.668.2412.329.3713.2210.76
BD-VAXThreat Model=Full Params2026.037.545.568.435.326.56.67