Human Preference Evaluation on Arena Creative Writing
23.4Win RateJS
Evaluation Results
| Method | Links | |
|---|---|---|
| JSBackbone=Llama-3.2-3B-Instruct, Reward Model=8B, Epochs=12025.11 | 23.4 | |
| RLOOBackbone=Llama-3.2-3B-Instruct, Reward Model=8B, Epochs=12025.11 | 20.2 | |
| Llama-3.2-3B-It2025.11 | 5.2 |