Sentiment Generation on IMDb (Average Score, Win Rate)
0.9882Average ScoreDPO*
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DPO*type=Oracle2026.02 | 0.9882 | 96.14 | |
| IPO*oracle=true2026.02 | 0.9788 | 96.44 | |
| DDPO2026.02 | 0.9748 | 93.3 | |
| DIPO+weighting=greater weight to human-labeled data2026.02 | 0.9428 | 89.58 | |
| DIPO2026.02 | 0.9414 | 87.44 | |
| IPO2026.02 | 0.9299 | 79.86 | |
| Sampled IPO2026.02 | 0.9153 | 86.84 | |
| DPOtraining_labels=mixed2026.02 | 0.8851 | 83.92 | |
| SFT2026.02 | 0.4662 | — | |
| SFT2026.02 | 0.4662 | — |