Information Completeness on QA-FEEDBACK (test)
23Win RateFINE-GRAINED RLHF
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| FINE-GRAINED RLHFCompared against=SFT2023.06 | 23 | 65.5 | 11.5 | |
| FINE-GRAINED RLHFCompared against=SFT-Full2023.06 | 22 | 61 | 17 | |
| FINE-GRAINED RLHFCompared against=Pref. RLHF2023.06 | 19.5 | 71 | 9.5 |