Question Answering on MAPLE-QA Qwen2.5-Omni-3B (eval)
57.24V ScoreMAPO
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| MAPOstrategy=adpw + cur2026.02 | 57.24 | 65.87 | 59.46 | 59.37 | 61.63 | 58.81 | 58.13 | 59.73 | |
| MAPOstrategy=adpw + adpcur, description=Full adaptive MAPO2026.02 | 57.13 | 66.4 | 61.35 | 59.16 | 61.64 | 58.33 | 58.67 | 59.82 | |
| MAPOstrategy=adpcur, curriculum=KL-based dynamic curriculum adaptation2026.02 | 56.87 | 65.33 | 61.89 | 58.62 | 61 | 59.52 | 58.27 | 59.38 | |
| MAPOstrategy=cur, curriculum=static curriculum2026.02 | 56.4 | 65.78 | 60.53 | 58.05 | 60.83 | 59.65 | 57.96 | 59.05 | |
| MAPOstrategy=Base2026.02 | 55.79 | 65.78 | 61.4 | 57.85 | 60.26 | 59.06 | 57.76 | 58.68 | |
| MAPOstrategy=adpw, weighting=batch weighting by historical KL divergence2026.02 | 55.34 | 67.47 | 62.16 | 58.28 | 60.77 | 58.57 | 57.96 | 58.98 | |
| MUPOmode=Modality-unaware baseline2026.02 | 55.08 | 65.34 | 63.82 | 57.47 | 60.15 | 58.77 | 58.14 | 58.58 | |
| Zero-shotmode=Zero-shot2026.02 | 34.03 | 33.6 | 42.16 | 38.58 | 41.4 | 43.45 | 40.86 | 39.78 |