Safety Evaluation on Harmful Fine-tuning Attacks Average
91.36ASRSFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SFTBase Model=LLaMA-3.2-3B-Instruct, Defense Method=SFT2026.05 | 91.36 | 4.39 | |
| SafeInstrBase Model=LLaMA-3.2-3B-Instruct, Defense Method=SafeInstr2026.05 | 82.01 | 4.04 | |
| PTSTBase Model=LLaMA-3.2-3B-Instruct, Defense Method=PTST2026.05 | 79.48 | 3.87 | |
| SafeGradBase Model=LLaMA-3.2-3B-Instruct, Defense Method=SafeGrad2026.05 | 71.77 | 3.54 | |
| LLaMA-3.2-3B-InstructBase Model=LLaMA-3.2-3B-Instruct2026.05 | 30.6 | 1.88 | |
| LisaBase Model=LLaMA-3.2-3B-Instruct, Defense Method=Lisa2026.05 | 24.19 | 1.73 | |
| SPARDBase Model=LLaMA-3.2-3B-Instruct, Defense Method=SPARD2026.05 | 12.09 | 1.38 |