Reasoning on Wino (leave-one-out setup)
86.2Accuracy (Wino LOO)Mashup Learning
Evaluation results
| Method | Links | |
|---|---|---|
| Mashup LearningModel=Gemma-3 4B, Setup=LoRA2026.03 | 86.2 | |
| From scratchModel=Gemma-3 4B, Setup=LoRA2026.03 | 85.9 | |
| Mashup LearningModel=Gemma-2 2B, Setup=LoRA2026.03 | 84.4 | |
| From scratchModel=Gemma-2 2B, Setup=LoRA2026.03 | 83.6 | |
| Mashup LearningModel=Gemma-3 4B, Setup=Full FT2026.03 | 83.3 | |
| From scratchModel=Gemma-3 4B, Setup=Full FT2026.03 | 81.9 | |
| Mashup LearningModel=Gemma-2 2B, Setup=Full FT2026.03 | 78.9 | |
| From scratchModel=Gemma-2 2B, Setup=Full FT2026.03 | 78.6 | |
| Mashup LearningModel=Gemma-3 1B, Setup=LoRA2026.03 | 71.2 | |
| From scratchModel=Gemma-3 1B, Setup=LoRA2026.03 | 70.6 | |
| Mashup LearningModel=Gemma-3 1B, Setup=Full FT2026.03 | 69.6 | |
| From scratchModel=Gemma-3 1B, Setup=Full FT2026.03 | 69.2 |