Mathematical Reasoning on MATH (Accuracy @ t1, t2 Comparison)
38.4Accuracy @ t1Prompt based
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Prompt basedBackbone=Gemma-2-27B-it, Configuration=w/o reflection2025.05 | 38.4 | 43.8 | 5.4 | |
| Prompt basedBackbone=Gemma-2-27B-it, Configuration=w/ reflection2025.05 | 38.4 | 45 | 6.6 | |
| Prompt basedBackbone=Gemma-2-9B-it, Configuration=w/o reflection2025.05 | 34.6 | 40.4 | 5.8 | |
| Prompt basedBackbone=Gemma-2-9B-it, Configuration=w/ reflection2025.05 | 34.6 | 40 | 5.4 | |
| ReflectEvoBackbone=Gemma-2-9B-it, Configuration=one-stage w/ D+2025.05 | 34.6 | 40 | 5.4 | |
| ReflectEvoBackbone=Gemma-2-9B-it, Configuration=two-stage w/ D+2025.05 | 34.6 | 35 | 0.4 | |
| ReflectEvoBackbone=Gemma-2-9B-it, Configuration=w/ D-2025.05 | 34.6 | 40 | 5.4 | |
| ReflectEvoBackbone=Gemma-2-9B-it, Configuration=w/ Dpref2025.05 | 34.6 | 40 | 5.4 | |
| SFT basedBackbone=Gemma-2-9B-it, Configuration=w/ SFT qa pairs2025.05 | 18.6 | — | — |