Direct Preference Optimization on SHP AlpacaEval 2.0
18.44LCWRFull Set
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Full SetSelection Method=Full Set2025.08 | 18.44 | 16.82 | |
| Difficulty-Based Preference Data SelectionSelection Method=Ours2025.08 | 18.41 | 19.52 | |
| Difficulty-Based Preference Data SelectionSelection Method=Ours2025.08 | 17.92 | 16.74 | |
| Full SetSelection Method=Full Set2025.08 | 17.84 | 16.52 | |
| ZIP†Selection Method=ZIP2025.08 | 17.83 | 16.77 | |
| RandomSelection Method=Random2025.08 | 17.53 | 17.49 | |
| ZIP†Selection Method=ZIP2025.08 | 17.22 | 16.03 | |
| DiverseEvol†Selection Method=DiverseEvol2025.08 | 17.2 | 16.59 | |
| DiverseEvol†Selection Method=DiverseEvol2025.08 | 16.98 | 15.77 | |
| SDPOSelection Method=SDPO2025.08 | 16.69 | 15.74 | |
| RandomSelection Method=Random2025.08 | 16.58 | 15.49 | |
| SDPOSelection Method=SDPO2025.08 | 16.58 | 15.96 | |
| Tulu3-SFTSelection Method=SFT Baseline2025.08 | 2.57 | 2.16 | |
| Tulu3-SFTSelection Method=SFT Baseline2025.08 | 2.57 | 2.16 |