List your company

Backdoor Defense on Refusal behavior dataset

80.3CACC (BadNet)Fine-mixing
22.0637.1852.367.42Jan 7, 2026

Evaluation results

MethodLinks
Fine-mixingBackbone=Qwen3-1.7B2026.0180.30.577.16177.532.678.42.3
Inst_cleanBackbone=Qwen3-1.7B2026.0179.80.579.83.279.81.479.80.9
MB-DefenseBackbone=Qwen3-1.7B2026.0177.30.973.42.375.70.577.10.5
ONIONBackbone=Qwen3-1.7B2026.0176.624.377.174.867.980.370.120.6
Clean-FFTBackbone=Qwen3-1.7B2026.0175.74.174.87874.368.379.42.3
Inst_atkBackbone=Qwen3-1.7B2026.0173.47.877.572.566.571.673.112.4
MB-DefenseBackbone=Llama3.2-1B2026.0139.4038.10.937.2036.20
Inst_cleanBackbone=Llama3.2-1B2026.01391.4391.4391.8391.8
Fine-mixingBackbone=Llama3.2-1B2026.0132.119.33377.131.246.827.119.3
Clean-FFTBackbone=Llama3.2-1B2026.0129.440.433.98333.970.626.634.9
ONIONBackbone=Llama3.2-1B2026.0125.251.433.979.432.679.42251.8
Inst_atkBackbone=Llama3.2-1B2026.0124.35535.387.628.478.918.344