Safety Evaluation on XSTEST (HS, ASR)
2.05HS RateTRIDENT-EDGE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TRIDENT-EDGEBackbone Model=META-LLAMA-3.1-8B2025.05 | 2.05 | 0.09 | |
| WILDBREAKBackbone Model=META-LLAMA-3.1-8B2025.05 | 2.1 | 0.07 | |
| AARTBackbone Model=META-LLAMA-3.1-8B2025.05 | 2.13 | 0.12 | |
| WILDCHATBackbone Model=META-LLAMA-3.1-8B2025.05 | 2.22 | 0.1 | |
| ATTAQBackbone Model=META-LLAMA-3.1-8B2025.05 | 2.31 | 0.15 | |
| SAFE_RLHFBackbone Model=META-LLAMA-3.1-8B2025.05 | 2.59 | 0.09 | |
| UnalignedBackbone Model=META-LLAMA-3.1-8B2025.05 | 2.77 | 0.32 | |
| HH_RLHFBackbone Model=META-LLAMA-3.1-8B2025.05 | 3.08 | 0.33 |