General Chat on Arena Hard v0.1
46.5Win RatePMLE
Evaluation Results
| Method | Links | |
|---|---|---|
| PMLEBackbone=LLaMA-3-8B-Instruct, Training Dataset=UltraFeedback2025.06 | 46.5 | |
| DPOBackbone=LLaMA-3-8B-Instruct, Training Dataset=UltraFeedback2025.06 | 45.1 | |
| Pref. Distill.Backbone=LLaMA-3-8B-Instruct, Training Dataset=UltraFeedback2025.06 | 41.9 | |
| REBELBackbone=LLaMA-3-8B-Instruct, Training Dataset=UltraFeedback2025.06 | 39.8 | |
| Base (LLaMA-3-8B-Instruct)Model Type=Instruct-only Base2025.06 | 26.8 |