Model Merging Performance Aggregation on MMMU, MME, SEED-Bench, OCRBench, TextVQA, OKVQA, GQA, and VizWiz
14.28Total Performance Gain SumAdaMMS
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| AdaMMSUnsupervised=true2025.03 | 14.28 | 7 | 416.67 | |
| Task ArithmeticUnsupervised=false2025.03 | 11.17 | 5 | 413.56 | |
| DARE-LinearUnsupervised=false2025.03 | 11.08 | 2 | 413.47 | |
| MetaGPTUnsupervised=true2025.03 | 11 | 4 | 413.39 | |
| AdaMMS2025.03 | 0.9192 | 9 | — | |
| Task Arithmetic2025.03 | 0.4745 | 7 | — | |
| MetaGPT2025.03 | -0.3694 | 2 | — | |
| DARE-Linear2025.03 | -0.4353 | 0 | — | |
| Ties-Merging2025.03 | -1.6805 | 0 | — | |
| DARE-Ties2025.03 | -2.6676 | 0 | — | |
| Ties-MergingUnsupervised=false2025.03 | -11.25 | 0 | 391.14 | |
| DARE-TiesUnsupervised=false2025.03 | -22.64 | 0 | 379.75 | |
| LLaVA(base)2025.03 | — | — | 409.29 | |
| mPLUG-Owl22025.03 | — | — | 395.5 |