Reasoning on Hella leave-one-out setup
94.7Average AccuracyMashup Learning
Evaluation Results
| Method | Links | |
|---|---|---|
| Mashup LearningModel=Gemma-3 4B, Setup=LoRA2026.03 | 94.7 | |
| From scratchModel=Gemma-3 4B, Setup=LoRA2026.03 | 94.3 | |
| Mashup LearningModel=Gemma-2 2B, Setup=LoRA2026.03 | 93.6 | |
| From scratchModel=Gemma-2 2B, Setup=LoRA2026.03 | 93.5 | |
| Mashup LearningModel=Gemma-3 4B, Setup=Full FT2026.03 | 92.7 | |
| From scratchModel=Gemma-3 4B, Setup=Full FT2026.03 | 92.6 | |
| Mashup LearningModel=Gemma-2 2B, Setup=Full FT2026.03 | 91.7 | |
| From scratchModel=Gemma-2 2B, Setup=Full FT2026.03 | 91.1 | |
| Mashup LearningModel=Gemma-3 1B, Setup=LoRA2026.03 | 82.7 | |
| From scratchModel=Gemma-3 1B, Setup=LoRA2026.03 | 81.4 | |
| Mashup LearningModel=Gemma-3 1B, Setup=Full FT2026.03 | 80.2 | |
| From scratchModel=Gemma-3 1B, Setup=Full FT2026.03 | 79.8 |