Safety Alignment on Safety Alignment Dataset 4-order (test)
100DSRMOSAIC-2
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MOSAIC-2Model=Llama-3.1-8B, # Params=8.2k/C2026.03 | 100 | 5.9 | |
| MOSAIC-5Model=Llama-3.1-8B, # Params=20.5k/C2026.03 | 100 | 1.8 | |
| MOSAIC-5Model=Llama-3.2-3B, # Params=15.4k/C2026.03 | 99.9 | 2.9 | |
| MOSAIC-2Model=Llama-3.2-3B, # Params=6.1k/C2026.03 | 99.7 | 5 | |
| SFTModel=Llama-3.1-8B, # Params=33.6M2026.03 | 98.9 | 6.1 | |
| SFTModel=Llama-3.2-3B, # Params=18.4M2026.03 | 98.9 | 5.2 | |
| ORPOModel=Llama-3.1-8B, # Params=33.6M2026.03 | 76.4 | 28.7 | |
| ORPOModel=Llama-3.2-3B, # Params=18.4M2026.03 | 72.9 | 29.1 | |
| In-contextModel=Llama-3.1-8B, # Params=0.4M/C2026.03 | 44.5 | 13.9 | |
| In-contextModel=Llama-3.2-3B, # Params=0.3M/C2026.03 | 41.9 | 12.3 |