List your company

Harmful Query on SORRY

12.79ASR (%)PeCAN
9.738430.336750.93571.5333Dec 12, 2025

Evaluation results

MethodLinks
PeCANBackbone=Llama3-8B-Instruct2025.1212.79
DPO-SBackbone=Llama3-8B-Instruct2025.1214.83
W-DOORBackbone=Qwen2.5-7B-Instruct2025.1215
NSPOBackbone=Llama3-8B-Instruct2025.1216.81
NSPOBackbone=Qwen2.5-7B-Instruct2025.1219.54
DPO-MixBackbone=Llama3-8B-Instruct2025.1223.26
BFPOBackbone=Llama3-8B-Instruct2025.1229.73
W-DOORBackbone=Llama3-8B-Instruct2025.1230.46
Llama3-8B-InstructBackbone=Llama3-8B-Instruct2025.1234.15
BFPOBackbone=Qwen2.5-7B-Instruct2025.1234.31
MoCANBackbone=Llama3-8B-Instruct2025.1236.45
SafeRLHFBackbone=Qwen2.5-7B-Instruct2025.1237.95
DPO-SBackbone=Qwen2.5-7B-Instruct2025.1240.91
DPO-HBackbone=Qwen2.5-7B-Instruct2025.1241.21
MoCANBackbone=Qwen2.5-7B-Instruct2025.1241.33
DPO-MixBackbone=Qwen2.5-7B-Instruct2025.1241.48
Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct2025.1241.88
SafeRLHFBackbone=Llama3-8B-Instruct2025.1244.77
PeCANBackbone=Qwen2.5-7B-Instruct2025.1245.42
DPO-HBackbone=Llama3-8B-Instruct2025.1289.08