Textual understanding on Arena-Hard
67.1Win RateUni-DPO
Evaluation Results
| Method | Links | |
|---|---|---|
| Uni-DPOModel=Gemma-2-9B-IT2025.06 | 67.1 | |
| SimPOModel=Gemma-2-9B-IT2025.06 | 59.1 | |
| Uni-DPOModel=Qwen2.5-7B2025.06 | 43.5 | |
| SFTModel=Gemma-2-9B-IT2025.06 | 40.8 | |
| Uni-DPO v0.2Model=Llama3-8B Instruct2025.06 | 40.6 | |
| SimPOModel=Qwen2.5-7B2025.06 | 39.5 | |
| Uni-DPOModel=Llama3-8B Instruct2025.06 | 37.3 | |
| SimPO v0.2Model=Llama3-8B Instruct2025.06 | 36.5 | |
| SimPOModel=Llama3-8B Instruct2025.06 | 33.8 | |
| DPOModel=Llama3-8B Instruct2025.06 | 32.6 | |
| Uni-DPO v0.2Model=Llama3-8B Base2025.06 | 30.7 | |
| SimPO v0.2Model=Llama3-8B Base2025.06 | 29.3 | |
| SFTModel=Llama3-8B Instruct2025.06 | 25.7 | |
| Uni-DPOModel=Llama3-8B Base2025.06 | 23.9 | |
| SimPOModel=Llama3-8B Base2025.06 | 23.4 | |
| DPOModel=Llama3-8B Base2025.06 | 15.9 | |
| SFTModel=Llama3-8B Base2025.06 | 3.3 |