Mathematical Reasoning on GSM8K (Accuracy, Overall Average Accuracy)
67.65AccuracyStandard FT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Standard FTModel=Gemma-3-4B2026.06 | 67.65 | 55.93 | |
| AlphaTokenModel=Gemma-3-4B2026.06 | 66.18 | 57.85 | |
| LESSModel=Gemma-3-4B2026.06 | 65.42 | 56.67 | |
| STMModel=Gemma-3-4B2026.06 | 64.15 | 56.24 | |
| ssTOKENModel=Gemma-3-4B2026.06 | 63.6 | 55.82 | |
| Token CleaningModel=Gemma-3-4B2026.06 | 63.2 | 55.57 | |
| LoRAModel=Gemma-3-4B2026.06 | 62.45 | 54.94 | |
| XTFModel=Gemma-3-4B2026.06 | 61.92 | 54.61 | |
| Pre-trainedModel=Gemma-3-4B2026.06 | 36.96 | 43.96 |