Harmlessness on Harmlessness
96Average Win RateCurriculum-RLAIF
Evaluation Results
| Method | Links | |
|---|---|---|
| Curriculum-RLAIFBase Model=Qwen2.5-32B, Category=Curriculum2025.05 | 96 | |
| Implicit Eval. (DPO)Base Model=Qwen2.5-32B, Category=Curriculum2025.05 | 94 | |
| Curriculum-RLAIFBase Model=LLaMA-3-8B, Category=Curriculum2025.05 | 93 | |
| Internal Eval.Base Model=Qwen2.5-32B, Category=Curriculum2025.05 | 93 | |
| Curriculum-RLAIFBase Model=Gemma-1-2B, Category=Curriculum2025.05 | 92 | |
| Conventional RLAIFBase Model=Qwen2.5-32B, Category=Non-Curriculum2025.05 | 91 | |
| Internal Eval.Base Model=Gemma-1-2B, Category=Curriculum2025.05 | 90 | |
| Implicit Eval. (DPO)Base Model=LLaMA-3-8B, Category=Curriculum2025.05 | 90 | |
| External Eval.Base Model=Qwen2.5-32B, Category=Curriculum2025.05 | 90 | |
| Internal Eval.Base Model=LLaMA-3-8B, Category=Curriculum2025.05 | 89 | |
| RLCDBase Model=Qwen2.5-32B, Category=Non-Curriculum2025.05 | 89 | |
| External Eval.Base Model=Gemma-1-2B, Category=Curriculum2025.05 | 88 | |
| Conventional RLAIFBase Model=LLaMA-3-8B, Category=Non-Curriculum2025.05 | 88 | |
| CAIBase Model=Qwen2.5-32B, Category=Non-Curriculum2025.05 | 88 | |
| Implicit Eval. (DPO)Base Model=Gemma-1-2B, Category=Curriculum2025.05 | 86 | |
| RLCDBase Model=LLaMA-3-8B, Category=Non-Curriculum2025.05 | 85 | |
| External Eval.Base Model=LLaMA-3-8B, Category=Curriculum2025.05 | 85 | |
| RLCDBase Model=Gemma-1-2B, Category=Non-Curriculum2025.05 | 83 | |
| Conventional RLAIFBase Model=Gemma-1-2B, Category=Non-Curriculum2025.05 | 83 | |
| CAIBase Model=LLaMA-3-8B, Category=Non-Curriculum2025.05 | 83 | |
| CAIBase Model=Gemma-1-2B, Category=Non-Curriculum2025.05 | 79 |