Reasoning on PIQA leave-one-out setup
88.5Average AccuracyMashup Learning
Evaluation results
| Method | Links | |
|---|---|---|
| Mashup LearningModel=Gemma-3 4B, Setup=LoRA2026.03 | 88.5 | |
| From scratchModel=Gemma-3 4B, Setup=LoRA2026.03 | 87.9 | |
| Mashup LearningModel=Gemma-2 2B, Setup=LoRA2026.03 | 86.3 | |
| From scratchModel=Gemma-3 4B, Setup=Full FT2026.03 | 86.1 | |
| Mashup LearningModel=Gemma-3 4B, Setup=Full FT2026.03 | 85.9 | |
| From scratchModel=Gemma-2 2B, Setup=LoRA2026.03 | 85.3 | |
| Mashup LearningModel=Gemma-2 2B, Setup=Full FT2026.03 | 83.1 | |
| From scratchModel=Gemma-2 2B, Setup=Full FT2026.03 | 83 | |
| Mashup LearningModel=Gemma-3 1B, Setup=LoRA2026.03 | 78.4 | |
| Mashup LearningModel=Gemma-3 1B, Setup=Full FT2026.03 | 77.7 | |
| From scratchModel=Gemma-3 1B, Setup=LoRA2026.03 | 77 | |
| From scratchModel=Gemma-3 1B, Setup=Full FT2026.03 | 76.1 |