Alignment on HH (IDN 20%)
78.2Win Rateα-UDPO
Evaluation Results
| Method | Links | |
|---|---|---|
| α-UDPOBase Model=Qwen-3-1.7B2026.07 | 78.2 | |
| UDPOBase Model=Qwen-3-1.7B2026.07 | 77.9 | |
| rDPOBase Model=Qwen-3-1.7B2026.07 | 73.3 | |
| DPOBase Model=Qwen-3-1.7B2026.07 | 68.7 | |
| α-URMBase Model=Qwen-3-1.7B2026.07 | 60.2 | |
| URMBase Model=Qwen-3-1.7B2026.07 | 59.5 | |
| rRMBase Model=Qwen-3-1.7B2026.07 | 58 | |
| RMBase Model=Qwen-3-1.7B2026.07 | 56.8 |