Personalized Response Generation on StereoSet Explicit Preference (test)
79.1Preference ScoreTriAlign
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| TriAlign2026.06 | 79.1 | 78.8 | 77.3 | 1 | 1.5 | 1.3 | |
| SWIFTYear=20262026.06 | 76.5 | 69.5 | 66.1 | 3 | 3.4 | 4.3 | |
| SFTYear=20222026.06 | 75.3 | 63.3 | 59.6 | 2.9 | 3.7 | 4.6 | |
| Qwen3-4B-Instruct-25072026.06 | 74.4 | 31.7 | 21.7 | 5.6 | 10 | 17.6 | |
| P-DEFENSEYear=20252026.06 | 74 | 37.2 | 32.1 | 2.6 | 5.1 | 6.9 | |
| BESTPERSONAYear=20242026.06 | 73.9 | 37.8 | 35.7 | 1.2 | 2 | 3.2 | |
| P-DEBIASYear=20242026.06 | 72.5 | 35.8 | 32.1 | 2.2 | 3.7 | 6 | |
| 2STEPPROMPTYear=20252026.06 | 45.8 | 33.5 | 31.9 | 0.9 | 1.6 | 2.6 |