Safety Evaluation on Xs (test)
100Safety ScoreREFLECTOR (+SFT)
Evaluation Results
| Method | Links | |
|---|---|---|
| REFLECTOR (+SFT)Backbone=Llama-3.1-8B-Instruct2026.05 | 100 | |
| REFLECTOR (+GDPO)Backbone=Llama-3.1-8B-Instruct2026.05 | 100 | |
| REFLECTOR (SFT)Backbone=Qwen-2.5-7B-Instruct2026.05 | 100 | |
| REFLECTOR (+GDPO)Backbone=Qwen-2.5-7B-Instruct2026.05 | 100 | |
| STAIRBackbone=Llama-3.1-8B-Instruct2026.05 | 99 | |
| STAIRBackbone=Qwen-2.5-7B-Instruct2026.05 | 99 | |
| Shallow-AlignBackbone=Llama-3.1-8B-Instruct2026.05 | 96.5 | |
| Shallow-AlignBackbone=Qwen-2.5-7B-Instruct2026.05 | 96 | |
| SFTBackbone=Llama-3.1-8B-Instruct2026.05 | 94.5 | |
| Self-CritiqueBackbone=Llama-3.1-8B-Instruct2026.05 | 88.5 | |
| OriginalBackbone=Llama-3.1-8B-Instruct2026.05 | 88 | |
| DPOBackbone=Llama-3.1-8B-Instruct2026.05 | 86 | |
| SFTBackbone=Qwen-2.5-7B-Instruct2026.05 | 84.5 | |
| Self-CritiqueBackbone=Qwen-2.5-7B-Instruct2026.05 | 75.25 | |
| OriginalBackbone=Qwen-2.5-7B-Instruct2026.05 | 73.5 | |
| DPOBackbone=Qwen-2.5-7B-Instruct2026.05 | 69.5 |