Safety Alignment on CoSApien Game Development
0.975SafetyCSRM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CSRMOptimization Method=Reward Distillation2026.05 | 0.975 | 3.675 | 3.5 | |
| Llama-3.1-8B-InstOptimization Method=Reward Distillation2026.05 | 0.94 | 3.565 | 3.38 | |
| Llama-3.1-8B-InstOptimization Method=Reinforce++2026.05 | 0.869 | 3.59 | 3.096 | |
| CSRMOptimization Method=Reinforce++2026.05 | 0.797 | 4.719 | 3.71 | |
| LlamaGuard-3-8BOptimization Method=Reinforce++2026.05 | 0.709 | 4.481 | 3.134 | |
| LlamaGuard-3-8BOptimization Method=Reward Distillation2026.05 | 0.675 | 4.8 | 3.025 |