Safety Evaluation on LLaMA-2-7B-CHAT (test)
0.55Safety ScoreTRAP
Evaluation Results
| Method | Links | |
|---|---|---|
| TRAPalpha=0.90, merging_technique=TIES-merging, DARE=true2024.10 | 0.55 | |
| TRAPalpha=0.70, merging_technique=TIES-merging, DARE=true2024.10 | 0.54 | |
| MERGEPRINTalpha=0.90, merging_technique=TIES-merging, DARE=true2024.10 | 0.54 | |
| MERGEPRINTalpha=0.70, merging_technique=TIES-merging, DARE=true2024.10 | 0.53 | |
| TRAPalpha=0.50, merging_technique=TIES-merging, DARE=true2024.10 | 0.52 | |
| TRAPalpha=0.90, merging_technique=TIES-merging, DARE=false2024.10 | 0.51 | |
| MERGEPRINTalpha=0.90, merging_technique=TIES-merging, DARE=false2024.10 | 0.51 | |
| MERGEPRINTalpha=0.50, merging_technique=TIES-merging, DARE=true2024.10 | 0.51 | |
| TRAPalpha=0.90, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.5 | |
| TRAPalpha=0.90, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.5 | |
| TRAPalpha=0.70, merging_technique=TIES-merging, DARE=false2024.10 | 0.5 | |
| IFalpha=0.50, merging_technique=TIES-merging, DARE=true2024.10 | 0.5 | |
| MERGEPRINTalpha=0.90, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.5 | |
| MERGEPRINTalpha=0.90, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.5 | |
| TRAPalpha=0.50, merging_technique=TIES-merging, DARE=false2024.10 | 0.49 | |
| IFalpha=0.70, merging_technique=TIES-merging, DARE=true2024.10 | 0.49 | |
| MERGEPRINTalpha=0.70, merging_technique=TIES-merging, DARE=false2024.10 | 0.49 | |
| TRAPalpha=0.30, merging_technique=TIES-merging, DARE=true2024.10 | 0.48 | |
| IFalpha=0.10, merging_technique=TIES-merging, DARE=true2024.10 | 0.48 | |
| IFalpha=0.30, merging_technique=TIES-merging, DARE=true2024.10 | 0.48 | |
| IFalpha=0.90, merging_technique=TIES-merging, DARE=true2024.10 | 0.48 | |
| MERGEPRINTalpha=0.50, merging_technique=TIES-merging, DARE=false2024.10 | 0.48 | |
| IFalpha=0.90, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.47 | |
| IFalpha=0.90, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.47 | |
| IFalpha=0.90, merging_technique=TIES-merging, DARE=false2024.10 | 0.47 | |
| MERGEPRINTalpha=0.30, merging_technique=TIES-merging, DARE=true2024.10 | 0.47 | |
| TRAPalpha=0.70, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.46 | |
| TRAPalpha=0.70, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.46 | |
| TRAPalpha=0.10, merging_technique=TIES-merging, DARE=true2024.10 | 0.46 | |
| IFalpha=0.70, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.46 | |
| IFalpha=0.70, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.46 | |
| IFalpha=0.50, merging_technique=TIES-merging, DARE=false2024.10 | 0.46 | |
| IFalpha=0.70, merging_technique=TIES-merging, DARE=false2024.10 | 0.46 | |
| MERGEPRINTalpha=0.10, merging_technique=TIES-merging, DARE=true2024.10 | 0.46 | |
| TRAPalpha=0.30, merging_technique=TIES-merging, DARE=false2024.10 | 0.45 | |
| IFalpha=0.30, merging_technique=TIES-merging, DARE=false2024.10 | 0.45 | |
| MERGEPRINTalpha=0.70, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.45 | |
| MERGEPRINTalpha=0.70, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.45 | |
| MERGEPRINTalpha=0.30, merging_technique=TIES-merging, DARE=false2024.10 | 0.45 | |
| TRAPalpha=0.50, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.44 | |
| TRAPalpha=0.50, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.44 | |
| IFalpha=0.50, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.44 | |
| IFalpha=0.50, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.44 | |
| IFalpha=0.10, merging_technique=TIES-merging, DARE=false2024.10 | 0.44 | |
| MERGEPRINTalpha=0.50, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.44 | |
| MERGEPRINTalpha=0.50, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.44 | |
| MERGEPRINTalpha=0.10, merging_technique=TIES-merging, DARE=false2024.10 | 0.44 | |
| TRAPalpha=0.30, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.43 | |
| TRAPalpha=0.30, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.43 | |
| TRAPalpha=0.10, merging_technique=TIES-merging, DARE=false2024.10 | 0.43 | |
| IFalpha=0.10, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.43 | |
| IFalpha=0.30, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.43 | |
| IFalpha=0.10, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.43 | |
| IFalpha=0.30, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.43 | |
| MERGEPRINTalpha=0.30, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.43 | |
| MERGEPRINTalpha=0.30, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.43 | |
| TRAPalpha=0.10, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.42 | |
| TRAPalpha=0.10, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.42 | |
| MERGEPRINTalpha=0.10, merging_technique=Task Arithmetic, DARE=false2024.10 | 0.42 | |
| MERGEPRINTalpha=0.10, merging_technique=Task Arithmetic, DARE=true2024.10 | 0.42 |