Preference Alignment on 15,000 listwise rankings (test)
0.384BT ScoreHard Panel
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Hard Panelbackbone=Llama-3.1-8B, bootstrap_n=10002026.02 | 0.384 | 0.375 | |
| US-Repbackbone=Llama-3.1-8B, bootstrap_n=10002026.02 | 0.102 | 0.113 | |
| Soft Panelbackbone=Llama-3.1-8B, bootstrap_n=10002026.02 | 0.017 | 0.071 | |
| Full PRISMbackbone=Llama-3.1-8B, bootstrap_n=10002026.02 | -0.07 | 0.026 | |
| Basebackbone=Llama-3.1-8B, bootstrap_n=10002026.02 | -0.432 | -0.584 |