Safety Evaluation on Harmful Evaluation Queries
75.2Final HSSFT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SFTModel=LLaMA3-8B-Instruct2026.05 | 75.2 | — | — | |
| SFTModel=LLaMA3-8B-LAT2026.05 | 74 | — | — | |
| SFTModel=LLaMA3-8B-ReFAT2026.05 | 74 | — | — | |
| SFTModel=LLaMA2-13B-Chat2026.05 | 33.2 | — | — | |
| BaseModel=LLaMA3-8B-Instruct2026.05 | 19.2 | — | — | |
| BaseModel=LLaMA3-8B-ReFAT2026.05 | 16.2 | — | — | |
| Buffer-and-Reinforce (ours)Model=LLaMA3-8B-Instruct2026.05 | 8.4 | 88.2 | 4.3 | |
| BaseModel=LLaMA2-13B-Chat2026.05 | 6.4 | — | — | |
| Buffer-and-Reinforce (ours)Model=LLaMA2-13B-Chat2026.05 | 4.3 | 85.7 | 1.5 | |
| Buffer-and-Reinforce (ours)Model=LLaMA3-8B-ReFAT2026.05 | 3.5 | 87.3 | 2.2 | |
| Buffer-and-Reinforce (ours)Model=LLaMA3-8B-LAT2026.05 | 3.1 | 87.6 | 2.1 |