Reasoning on ARC-e (leave-one-out setup)
93.5Accuracy (ARC-e LOO)Mashup Learning
Evaluation results
| Method | Links | |
|---|---|---|
| Mashup LearningModel=Gemma-3 4B, Setup=LoRA2026.03 | 93.5 | |
| Mashup LearningModel=Gemma-3 4B, Setup=Full FT2026.03 | 92.3 | |
| Mashup LearningModel=Gemma-2 2B, Setup=LoRA2026.03 | 91.6 | |
| From scratchModel=Gemma-3 4B, Setup=LoRA2026.03 | 91.6 | |
| From scratchModel=Gemma-3 4B, Setup=Full FT2026.03 | 90.7 | |
| From scratchModel=Gemma-2 2B, Setup=LoRA2026.03 | 90.1 | |
| Mashup LearningModel=Gemma-2 2B, Setup=Full FT2026.03 | 90.1 | |
| From scratchModel=Gemma-2 2B, Setup=Full FT2026.03 | 89.3 | |
| Mashup LearningModel=Gemma-3 1B, Setup=LoRA2026.03 | 78.8 | |
| Mashup LearningModel=Gemma-3 1B, Setup=Full FT2026.03 | 78.1 | |
| From scratchModel=Gemma-3 1B, Setup=LoRA2026.03 | 75 | |
| From scratchModel=Gemma-3 1B, Setup=Full FT2026.03 | 73.9 |