Safety-Utility Trade-off on XS (test)
45.56XS Test ScoreDPO
Evaluation Results
| Method | Links | |
|---|---|---|
| DPOModel=Llama-3-8B2025.05 | 45.56 | |
| APLModel=Llama-3-8B2025.05 | 45.56 | |
| APL (RM)Model=Llama-3-8B2025.05 | 44.89 | |
| APL (w/o Adv)Model=Llama-3-8B2025.05 | 44.22 | |
| CAPOModel=Llama-3-8B2025.05 | 44.03 | |
| BaseModel=Llama-3-8B2025.05 | 44 | |
| APLModel=Mistral-7B2025.05 | 40.89 | |
| APL (RM)Model=Mistral-7B2025.05 | 38.44 | |
| APL (w/o Adv)Model=Mistral-7B2025.05 | 31.56 | |
| DPOModel=Mistral-7B2025.05 | 29.33 | |
| CAPOModel=Mistral-7B2025.05 | 28.67 | |
| BaseModel=Mistral-7B2025.05 | 20 |