Safety Alignment on CoSApien
99.5Safety ScoreLlama-3.1-8B-Inst
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Llama-3.1-8B-InstOptimization Method=Reward Distillation2026.05 | 99.5 | 3.43 | 3.405 | |
| CSRMOptimization Method=Reward Distillation2026.05 | 99 | 3.99 | 3.65 | |
| Llama-3.1-8B-InstOptimization Method=Reinforce++2026.05 | 93.1 | 3.044 | 2.831 | |
| CSRMOptimization Method=Reinforce++2026.05 | 82.5 | 4.706 | 3.884 | |
| LlamaGuard-3-8BOptimization Method=Reinforce++2026.05 | 70.9 | 4.481 | 3.134 | |
| LlamaGuard-3-8BOptimization Method=Reward Distillation2026.05 | 70.5 | 4.725 | 3.37 |