Mathematical Reasoning on MathQA (Accuracy, ECE, Brier)
80.2AccuracySFT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SFTBackbone Model=Ministral 3 8B2026.06 | 80.2 | 12.4 | 29.51 | |
| CoRABackbone Model=Ministral 3 8B2026.06 | 80.2 | 6.58 | 22.66 | |
| GRPOBackbone Model=Ministral 3 8B2026.06 | 79.8 | 7.88 | 25.86 | |
| BaseBackbone Model=Ministral 3 8B2026.06 | 79.4 | 8.35 | 26.43 | |
| GRPOBackbone Model=Qwen2.5 7B2026.06 | 77.2 | 13.94 | 35.5 | |
| CoRABackbone Model=Qwen2.5 7B2026.06 | 76.4 | 11.09 | 32.83 | |
| SFTBackbone Model=Qwen2.5 7B2026.06 | 72.8 | 17.55 | 41.69 | |
| BaseBackbone Model=Qwen2.5 7B2026.06 | 72 | 11.55 | 37.69 | |
| CoRABackbone Model=Gemma 2 9B Instruct2026.06 | 63.4 | 25.6 | 58.27 | |
| GRPOBackbone Model=Gemma 2 9B Instruct2026.06 | 61.8 | 29.9 | 64.7 | |
| SFTBackbone Model=Gemma 2 9B Instruct2026.06 | 61.2 | 34.43 | 70.67 | |
| BaseBackbone Model=Gemma 2 9B Instruct2026.06 | 58.8 | 32.21 | 68.08 |