Multi-turn Dialogue on Honor-Dialogue
88.13Life Services Domain PerformanceDVPO
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DVPOtraining steps=5002025.12 | 88.13 | 87.73 | 87.67 | 87.67 | 82.73 | 86.79 | |
| Reinforce++training steps=5002025.12 | 87.87 | 81.07 | 77.67 | 87.13 | 85.53 | 83.85 | |
| Baselinetraining steps=5002025.12 | 86.73 | 84.5 | 90.23 | 87.13 | 82.7 | 86.26 | |
| PPOtraining steps=5002025.12 | 84.2 | 86.07 | 90.87 | 81 | 83.87 | 85.2 | |
| Dr.GRPOtraining steps=5002025.12 | 51.67 | 60.3 | 61.37 | 52.17 | 56.47 | 56.39 | |
| GRPOtraining steps=5002025.12 | 30.03 | 28.57 | 28.33 | 28.9 | 27.9 | 28.75 |