Reasoning on MathQA leave-one-out setup
56.9Average AccuracyMashup Learning
Evaluation results
| Method | Links | |
|---|---|---|
| Mashup LearningModel=Gemma-3 4B, Setup=LoRA2026.03 | 56.9 | |
| From scratchModel=Gemma-3 4B, Setup=LoRA2026.03 | 54.9 | |
| Mashup LearningModel=Gemma-3 4B, Setup=Full FT2026.03 | 46.5 | |
| From scratchModel=Gemma-3 4B, Setup=Full FT2026.03 | 46.2 | |
| Mashup LearningModel=Gemma-2 2B, Setup=LoRA2026.03 | 45.8 | |
| From scratchModel=Gemma-2 2B, Setup=LoRA2026.03 | 44.7 | |
| Mashup LearningModel=Gemma-2 2B, Setup=Full FT2026.03 | 40.7 | |
| From scratchModel=Gemma-2 2B, Setup=Full FT2026.03 | 40.5 | |
| Mashup LearningModel=Gemma-3 1B, Setup=LoRA2026.03 | 39.8 | |
| From scratchModel=Gemma-3 1B, Setup=LoRA2026.03 | 39.3 | |
| Mashup LearningModel=Gemma-3 1B, Setup=Full FT2026.03 | 39.1 | |
| From scratchModel=Gemma-3 1B, Setup=Full FT2026.03 | 37.3 |