Safety Evaluation on ReNeLLM
97.05Safety ScoreREFLECTOR (+GDPO)
Evaluation results
| Method | Links | |
|---|---|---|
| REFLECTOR (+GDPO)Backbone=Llama-3.1-8B-Instruct2026.05 | 97.05 | |
| REFLECTOR (+GDPO)Backbone=Qwen-2.5-7B-Instruct2026.05 | 94.23 | |
| REFLECTOR (+SFT)Backbone=Llama-3.1-8B-Instruct2026.05 | 93.92 | |
| REFLECTOR (SFT)Backbone=Qwen-2.5-7B-Instruct2026.05 | 88.51 | |
| STAIRBackbone=Qwen-2.5-7B-Instruct2026.05 | 78.65 | |
| STAIRBackbone=Llama-3.1-8B-Instruct2026.05 | 77.27 | |
| Shallow-AlignBackbone=Qwen-2.5-7B-Instruct2026.05 | 73.4 | |
| Shallow-AlignBackbone=Llama-3.1-8B-Instruct2026.05 | 72.1 | |
| Self-CritiqueBackbone=Qwen-2.5-7B-Instruct2026.05 | 48.65 | |
| Self-CritiqueBackbone=Llama-3.1-8B-Instruct2026.05 | 48.5 | |
| DPOBackbone=Llama-3.1-8B-Instruct2026.05 | 47.5 | |
| DPOBackbone=Qwen-2.5-7B-Instruct2026.05 | 45.96 | |
| SFTBackbone=Llama-3.1-8B-Instruct2026.05 | 44 | |
| OriginalBackbone=Llama-3.1-8B-Instruct2026.05 | 42.7 | |
| SFTBackbone=Qwen-2.5-7B-Instruct2026.05 | 42.11 | |
| OriginalBackbone=Qwen-2.5-7B-Instruct2026.05 | 40.19 |