Discriminative Performance on Ultrafeedback 61.1k (test)
73.05AccuracyFA-DPO
Evaluation Results
| Method | Links | |
|---|---|---|
| FA-DPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.11 | 73.05 | |
| FA-DPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.11 | 70.97 | |
| FA-DPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.11 | 70.77 | |
| ROPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.11 | 70.75 | |
| rDPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.11 | 70.13 | |
| FA-DPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.11 | 69.77 | |
| DPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.11 | 68.22 | |
| cDPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.11 | 67.2 | |
| FA-DPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.11 | 67.2 | |
| rDPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.11 | 65.9 | |
| ROPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.11 | 64.93 | |
| SIMPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.11 | 63.73 | |
| SIMPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.11 | 63.64 | |
| cDPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.11 | 62.57 | |
| DPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.11 | 61.77 | |
| rDPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.11 | 61.7 | |
| cDPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.11 | 59 | |
| ROPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.11 | 58.93 | |
| DPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.11 | 58.63 | |
| SIMPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.11 | 57.81 | |
| rDPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.11 | 56.87 | |
| cDPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.11 | 56.67 | |
| SIMPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.11 | 55.53 | |
| DPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.11 | 55.43 | |
| ROPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.11 | 55.23 | |
| ROPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.11 | 54.07 | |
| cDPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.11 | 53.93 | |
| DPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.11 | 51.87 | |
| SIMPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.11 | 51.77 | |
| rDPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.11 | 47.67 |