Mathematical Reasoning on MATH (acc, len)
76.67Accuracy (MATH)GRPO-O1
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GRPO-O1Backbone=Gemma3-4B2025.09 | 76.67 | 834 | |
| GRPOBackbone=Gemma3-4B2025.09 | 76.54 | 1,177 | |
| GRPO-ERBackbone=Gemma3-4B2025.09 | 75.73 | 979 | |
| SA-GRPOBackbone=Gemma3-4B2025.09 | 75.55 | 1,130 | |
| Gemma3-4BBackbone=Gemma3-4B, Strategy=Base2025.09 | 68.33 | 837 | |
| GRPOBackbone=Qwen2-2B-VL2025.09 | 25.25 | 321 | |
| GRPO-ERBackbone=Qwen2-2B-VL2025.09 | 23.73 | 212 | |
| GRPO-O1Backbone=Qwen2-2B-VL2025.09 | 23.2 | 79 | |
| SA-GRPOBackbone=Qwen2-2B-VL2025.09 | 21.22 | 84 | |
| Qwen2-2B-VLBackbone=Qwen2-2B-VL, Strategy=Base2025.09 | 14.94 | 15.84 |