Visual Language Understanding on MME v1 (full)
1,657.9ScoreMAGIC
Evaluation Results
| Method | Links | |
|---|---|---|
| MAGICTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,657.9 | |
| COINCIDETraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,495.6 | |
| ICONSTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,485.7 | |
| RandomTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,483 | |
| FullTraining Data Fraction=100%, Backbone=LLaVA-1.5-7B2026.05 | 1,476.9 | |
| DataTailorTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,476.1 | |
| COIDOTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,450.2 | |
| EL2NTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,439.5 | |
| D2-PruningTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,391.2 | |
| SemDeDupTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,376.9 | |
| PerplexityTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,341.4 | |
| Self-SepTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,335.9 | |
| CLIP-ScoreTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,331.6 | |
| Self-FilterTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,306.2 | |
| RDSTraining Data Fraction=20%, Backbone=LLaVA-1.5-7B2026.05 | 1,093.8 |