Reasoning on CSQA (leave-one-out setup)
83.8AccuracyMashup Learning
Evaluation results
| Method | Links | |
|---|---|---|
| Mashup LearningModel=Gemma-3 4B, Setup=LoRA2026.03 | 83.8 | |
| From scratchModel=Gemma-3 4B, Setup=LoRA2026.03 | 83.2 | |
| Mashup LearningModel=Gemma-2 2B, Setup=LoRA2026.03 | 82.7 | |
| From scratchModel=Gemma-2 2B, Setup=LoRA2026.03 | 82.5 | |
| Mashup LearningModel=Gemma-3 4B, Setup=Full FT2026.03 | 80.7 | |
| From scratchModel=Gemma-3 4B, Setup=Full FT2026.03 | 80.6 | |
| Mashup LearningModel=Gemma-2 2B, Setup=Full FT2026.03 | 78.4 | |
| From scratchModel=Gemma-2 2B, Setup=Full FT2026.03 | 77.7 | |
| Mashup LearningModel=Gemma-3 1B, Setup=LoRA2026.03 | 73.5 | |
| From scratchModel=Gemma-3 1B, Setup=LoRA2026.03 | 71.9 | |
| Mashup LearningModel=Gemma-3 1B, Setup=Full FT2026.03 | 70.5 | |
| From scratchModel=Gemma-3 1B, Setup=Full FT2026.03 | 68.8 |