Mathematical Reasoning on MathVerse (test)
44.88AccuracyQwen2.5-VL-7B + Cont. Reward
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-VL-7B + Cont. Rewardreward_type=Continuous2025.11 | 44.88 | |
| Vision-Zeroexternal supervision=true2025.11 | 43.86 | |
| Qwen2.5-VL-7B (Baseline)2025.11 | 43.78 | |
| Qwen2.5-VL-7B + Discrete Rewardreward_type=Discrete2025.11 | 42.1 |