Visual Reasoning on Out-of-Domain (OOD) Aggregate (HalluBench, MathVista, MathVerse, MathVision)
0.5531OOD Avg AccuracySaEI
Evaluation Results
| Method | Links | |
|---|---|---|
| SaEIFinetuning status=Finetuned, Group size=n1 = n2 = 6 (total n = 12)2025.12 | 0.5531 | |
| KL-CovFinetuning status=Finetuned, Group size=n = 122025.12 | 0.5474 | |
| Vanilla GRPOFinetuning status=Finetuned, Group size=n = 122025.12 | 0.547 | |
| NoisyRolloutFinetuning status=Finetuned, Group size=n1 = n2 = 6 (total n = 12)2025.12 | 0.5368 | |
| Qwen2.5-VL-7B-InstructFinetuning status=Not finetuned2025.12 | 0.4999 |