Reward Modeling on Meta Helpfulness (test)
63.2AccuracyHelpfulness RM
Evaluation results
| Method | Links | |
|---|---|---|
| Helpfulness RMFine-tuned=true2023.07 | 63.2 | |
| GPT4zero-shot=true2023.07 | 58.6 | |
| Safety RMFine-tuned=true2023.07 | 56.2 | |
| Open AssistantBackbone=DeBERTa V3 Large2023.07 | 53.8 | |
| SteamSHP-XLBackbone=FLAN-T5-xl2023.07 | 52.8 |