List your company

Backdoor Attack Evaluation on StrongREJECT

0.601ASR (w/ trigger)DualEdit
-0.024040.138230.30050.46277Apr 14, 2026

Evaluation results

MethodLinks
DualEditBackbone=Llama-2-7b-chat2026.040.6011.40.299
STEEREDITBackbone=Llama-2-7b-chat2026.040.610.136
STEEREDITBackbone=Llama-3-8B-Instruct2026.040.5352.30.028
SFTBackbone=Qwen2.5-7B-Instruct2026.040.5160.80.164
STEEREDITBackbone=Qwen2.5-7B-Instruct2026.040.4576.30.111
SFTBackbone=Llama-3-8B-Instruct2026.040.4470.10.135
SFTBackbone=Llama-2-7b-chat2026.040.410.40.192
DualEditBackbone=Qwen2.5-7B-Instruct2026.040.3996.50.184
JailbreakEditBackbone=Llama-2-7b-chat2026.040.38510.426
DualEditBackbone=Llama-3-8B-Instruct2026.040.3722.40.087
JailbreakEditBackbone=Qwen2.5-7B-Instruct2026.040.3026.30.229
JailbreakEditBackbone=Llama-3-8B-Instruct2026.040.2672.10.292
ROMEBackbone=Llama-3-8B-Instruct2026.040.2653.50.076
ROMEBackbone=Llama-2-7b-chat2026.040.10300.352
ROMEBackbone=Qwen2.5-7B-Instruct2026.040.0994.50.387
Qwen2.5-7B-InstructModel Type=Base Model2026.040.0594.90.369
Llama-3-8B-InstructModel Type=Base Model2026.040.011.40.024
Llama-2-7b-chatModel Type=Base Model2026.0400.30.108