Safety and Utility Evaluation on Safety and Utility evaluation suite (test)
3HS ScoreBuffer-and-Reinforce
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Buffer-and-ReinforceBase Model Architecture=Gemma2-9B-it2026.05 | 3 | 81.4 | |
| LDIFSBase Model Architecture=Gemma2-9B-it2026.05 | 4.1 | 81 | |
| Buffer-and-ReinforceBase Model Architecture=Average2026.05 | 5.2 | 78.3 | |
| AntidoteBase Model Architecture=Gemma2-9B-it2026.05 | 5.3 | 81 | |
| AsFTBase Model Architecture=Gemma2-9B-it2026.05 | 5.5 | 77.4 | |
| Buffer-and-ReinforceBase Model Architecture=Qwen3-4B-Instruct2026.05 | 5.5 | 80 | |
| Security VectorBase Model Architecture=Gemma2-9B-it2026.05 | 6.5 | 81.2 | |
| LisaBase Model Architecture=Gemma2-9B-it2026.05 | 7.7 | 79.7 | |
| SafeLoRABase Model Architecture=Gemma2-9B-it2026.05 | 7.8 | 78.8 | |
| Buffer-and-ReinforceBase Model Architecture=LLaMA3-8B-Instruct2026.05 | 8.4 | 77.5 | |
| AntidoteBase Model Architecture=Qwen3-4B-Instruct2026.05 | 13 | 76.5 | |
| SafeLoRABase Model Architecture=Qwen3-4B-Instruct2026.05 | 14.2 | 56.8 | |
| AntidoteBase Model Architecture=Average2026.05 | 14.5 | 77.4 | |
| Security VectorBase Model Architecture=Average2026.05 | 15 | 78.2 | |
| Security VectorBase Model Architecture=Qwen3-4B-Instruct2026.05 | 16.9 | 81.1 | |
| LDIFSBase Model Architecture=LLaMA3-8B-Instruct2026.05 | 17.2 | 73.8 | |
| AsFTBase Model Architecture=Average2026.05 | 18.3 | 67.9 | |
| SafeInstructBase Model Architecture=Gemma2-9B-it2026.05 | 18.5 | 79.9 | |
| LisaBase Model Architecture=Average2026.05 | 19 | 76.6 | |
| SafeLoRABase Model Architecture=Average2026.05 | 20.7 | 79.4 | |
| SafeInstructBase Model Architecture=LLaMA3-8B-Instruct2026.05 | 21.5 | 69.8 | |
| Security VectorBase Model Architecture=LLaMA3-8B-Instruct2026.05 | 21.7 | 72.2 | |
| SafeInstructBase Model Architecture=Average2026.05 | 22.4 | 76.6 | |
| LisaBase Model Architecture=Qwen3-4B-Instruct2026.05 | 22.6 | 80.9 | |
| AsFTBase Model Architecture=Qwen3-4B-Instruct2026.05 | 23.5 | 58 | |
| AntidoteBase Model Architecture=LLaMA3-8B-Instruct2026.05 | 25.3 | 74.7 | |
| PanaceaBase Model Architecture=Qwen3-4B-Instruct2026.05 | 25.9 | 80.3 | |
| AsFTBase Model Architecture=LLaMA3-8B-Instruct2026.05 | 26 | 68.3 | |
| SafeLoRABase Model Architecture=LLaMA3-8B-Instruct2026.05 | 26 | 72.8 | |
| LisaBase Model Architecture=LLaMA3-8B-Instruct2026.05 | 26.8 | 69.1 | |
| SafeInstructBase Model Architecture=Qwen3-4B-Instruct2026.05 | 27.3 | 80 | |
| LDIFSBase Model Architecture=Average2026.05 | 30.8 | 78.2 | |
| PanaceaBase Model Architecture=Gemma2-9B-it2026.05 | 32 | 78.7 | |
| PanaceaBase Model Architecture=Average2026.05 | 33 | 74.8 | |
| PanaceaBase Model Architecture=LLaMA3-8B-Instruct2026.05 | 41 | 65.5 | |
| SFTBase Model Architecture=Gemma2-9B-it2026.05 | 59.1 | 80 | |
| SFTBase Model Architecture=Average2026.05 | 68.8 | 76 | |
| LDIFSBase Model Architecture=Qwen3-4B-Instruct2026.05 | 71 | 79.9 | |
| SFTBase Model Architecture=Qwen3-4B-Instruct2026.05 | 72.2 | 79.6 | |
| SFTBase Model Architecture=LLaMA3-8B-Instruct2026.05 | 75.2 | 68.4 |