Visual Reasoning on Geo3K
56.74AccuracyQwen2.5-VL-32B-Instruct + NoisyRollout
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-VL-32B-Instruct + NoisyRolloutStudent Model=Qwen2.5-VL-32B-Instruct, Distillation/Training Method=NoisyRollout2026.03 | 56.74 | |
| REOPOLDTeacher Model=Qwen2.5-VL-32B-Instruct + NoisyRollout, Student Model=Qwen2.5-VL-7B-Instruct, Distillation/Training Method=REOPOLD2026.03 | 53.58 | |
| RKLTeacher Model=Qwen2.5-VL-32B-Instruct + NoisyRollout, Student Model=Qwen2.5-VL-7B-Instruct, Distillation/Training Method=RKL2026.03 | 51.75 | |
| GRPOTeacher Model=Qwen2.5-VL-32B-Instruct + NoisyRollout, Student Model=Qwen2.5-VL-7B-Instruct, Distillation/Training Method=GRPO2026.03 | 51.4 | |
| REOPOLDTeacher Model=Qwen2.5-VL-32B-Instruct + NoisyRollout, Student Model=Qwen2.5-VL-3B-Instruct, Distillation/Training Method=REOPOLD2026.03 | 50.58 | |
| NoisyRolloutTeacher Model=Qwen2.5-VL-32B-Instruct + NoisyRollout, Student Model=Qwen2.5-VL-7B-Instruct, Distillation/Training Method=NoisyRollout2026.03 | 50.08 | |
| RKLTeacher Model=Qwen2.5-VL-32B-Instruct + NoisyRollout, Student Model=Qwen2.5-VL-3B-Instruct, Distillation/Training Method=RKL2026.03 | 48.09 | |
| Qwen2.5-VL-7B-InstructStudent Model=Qwen2.5-VL-7B-Instruct2026.03 | 39.77 | |
| PAPOTeacher Model=Qwen2.5-VL-32B-Instruct + NoisyRollout, Student Model=Qwen2.5-VL-3B-Instruct, Distillation/Training Method=PAPO2026.03 | 32.95 | |
| Qwen2.5-VL-3B-InstructStudent Model=Qwen2.5-VL-3B-Instruct2026.03 | 26.46 |