List your company

Sentiment Steering on LLaMA2-7B-Chat (test)

2.01BadNetsBD-VAX
-0.309615.347731.00546.6623Mar 13, 2026

Evaluation results

MethodLinks
BD-VAXThreat Model=Lora Adapter2026.032.010.510020.91
BD-VAXThreat Model=Full Params2026.032.511.0100.56.52.1
Fine-PruningThreat Model=Lora Adapter2026.0313.573.530013.56.12
Fine-PruningThreat Model=Full Params2026.0318.591.010.511.0129.510.94
CROWThreat Model=Full Params2026.0321.113.080.50.518.098.66
FTThreat Model=Lora Adapter2026.0323.59003.0124.510.62
QuantizationThreat Model=Full Params2026.0331.552439.2916.36
PruningThreat Model=Full Params2026.0336.341.514.520.613.78
PruningThreat Model=Lora Adapter2026.0347.479.022.532.0850.4822.32
No DefenseThreat Model=N/A2026.0359.313.684.33.526028.16
FTThreat Model=Full Params2026.036013.755.083.5263.4729.96