Accuracy on TruthfulQA (TQA) (Accuracy & Drop)
66.4Accuracy (TQA)UDPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| UDPOBase Model=Qwen, Evaluation Setting=Under attack2026.02 | 66.4 | -14.8 | |
| RMAACBase Model=Qwen, Evaluation Setting=Under attack2026.02 | 44.8 | -31.4 | |
| MAPPOBase Model=Qwen, Evaluation Setting=Under attack2026.02 | 36.1 | -39.3 | |
| Standard MADBase Model=Qwen, Evaluation Setting=Under attack2026.02 | 28.4 | -44.1 |