Human Preference Evaluation on Arena-Hard v0.1
56.7Win RateJS
Evaluation Results
| Method | Links | |
|---|---|---|
| JSBackbone=Llama-3.2-3B-Instruct, Reward Model=8B, Epochs=12025.11 | 56.7 | |
| RLOOBackbone=Llama-3.2-3B-Instruct, Reward Model=8B, Epochs=12025.11 | 55.3 | |
| Llama-3.2-3B-It2025.11 | 26.2 |