List your company

Safety Evaluation on AutoDAN

100Safety ScoreREFLECTOR (+GDPO)
35.998452.614269.2385.8458May 20, 2026

Evaluation results

MethodLinks
REFLECTOR (+GDPO)Backbone=Llama-3.1-8B-Instruct2026.05100
STAIRBackbone=Llama-3.1-8B-Instruct2026.0599.04
REFLECTOR (+SFT)Backbone=Llama-3.1-8B-Instruct2026.0598.26
REFLECTOR (+GDPO)Backbone=Qwen-2.5-7B-Instruct2026.0597.89
Shallow-AlignBackbone=Llama-3.1-8B-Instruct2026.0596.8
Self-CritiqueBackbone=Llama-3.1-8B-Instruct2026.0596.15
DPOBackbone=Llama-3.1-8B-Instruct2026.0595.38
STAIRBackbone=Qwen-2.5-7B-Instruct2026.0595.19
OriginalBackbone=Llama-3.1-8B-Instruct2026.0594.75
SFTBackbone=Llama-3.1-8B-Instruct2026.0594.62
Shallow-AlignBackbone=Qwen-2.5-7B-Instruct2026.0591.8
REFLECTOR (SFT)Backbone=Qwen-2.5-7B-Instruct2026.0590.38
OriginalBackbone=Qwen-2.5-7B-Instruct2026.0552.89
DPOBackbone=Qwen-2.5-7B-Instruct2026.0549.04
Self-CritiqueBackbone=Qwen-2.5-7B-Instruct2026.0543.7
SFTBackbone=Qwen-2.5-7B-Instruct2026.0538.46