List your company

Safety Evaluation on GCG

95.96Safety ScoreREFLECTOR (+GDPO)
38.593653.486868.3883.2732May 20, 2026

Evaluation results

MethodLinks
REFLECTOR (+GDPO)Backbone=Qwen-2.5-7B-Instruct2026.0595.96
REFLECTOR (+GDPO)Backbone=Llama-3.1-8B-Instruct2026.0594.23
REFLECTOR (+SFT)Backbone=Llama-3.1-8B-Instruct2026.0590.96
REFLECTOR (SFT)Backbone=Qwen-2.5-7B-Instruct2026.0588.69
STAIRBackbone=Qwen-2.5-7B-Instruct2026.0587.69
STAIRBackbone=Llama-3.1-8B-Instruct2026.0586.15
Shallow-AlignBackbone=Qwen-2.5-7B-Instruct2026.0583.5
Shallow-AlignBackbone=Llama-3.1-8B-Instruct2026.0583.4
DPOBackbone=Llama-3.1-8B-Instruct2026.0582.31
Self-CritiqueBackbone=Llama-3.1-8B-Instruct2026.0581.15
SFTBackbone=Llama-3.1-8B-Instruct2026.0579.8
OriginalBackbone=Llama-3.1-8B-Instruct2026.0578.4
DPOBackbone=Qwen-2.5-7B-Instruct2026.0551.92
Self-CritiqueBackbone=Qwen-2.5-7B-Instruct2026.0546.8
SFTBackbone=Qwen-2.5-7B-Instruct2026.0542.31
OriginalBackbone=Qwen-2.5-7B-Instruct2026.0540.8