Generative Performance on HH Golden 42.5k (test)
88.9Win RateFA-DPO
Evaluation Results
| Method | Links | |
|---|---|---|
| FA-DPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.11 | 88.9 | |
| FA-DPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.11 | 88.7 | |
| cDPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.11 | 88.35 | |
| FA-DPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.11 | 87.7 | |
| DPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.11 | 87 | |
| FA-DPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.11 | 82.85 | |
| rDPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.11 | 82.8 | |
| ROPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.11 | 81.15 | |
| FA-DPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.11 | 78.6 | |
| rDPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.11 | 77.3 | |
| ROPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.11 | 70.5 | |
| SIMPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.11 | 65.1 | |
| cDPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.11 | 62.3 | |
| rDPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.11 | 58.1 | |
| cDPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.11 | 56.95 | |
| ROPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.11 | 54.3 | |
| rDPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.11 | 54.1 | |
| rDPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.11 | 53.55 | |
| ROPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.11 | 53.5 | |
| ROPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.11 | 49.2 | |
| cDPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.11 | 49.05 | |
| cDPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.11 | 37.75 | |
| DPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.11 | 34.05 | |
| SIMPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.11 | 29.95 | |
| DPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.11 | 29.95 | |
| SIMPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.11 | 27.5 | |
| DPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.11 | 26.85 | |
| SIMPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.11 | 24.9 | |
| SIMPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.11 | 22.4 | |
| DPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.11 | 21.05 |