Multi-task learning evaluation on MM-MergeBench 20 tasks 1.0 (test)
91.37Average Absolute AccuracyIndividual
Evaluation results
| Method | Links | ||
|---|---|---|---|
| IndividualBackbone=ViT-B-32, Setting=FFT2026.03 | 91.37 | — | |
| DC-MergeBackbone=ViT-B-32, Setting=FFT2026.03 | 80.58 | 88.18 | |
| Iso-CTSBackbone=ViT-B-32, Setting=FFT2026.03 | 78.05 | 85.48 | |
| TSV-MBackbone=ViT-B-32, Setting=FFT2026.03 | 77.07 | 84.29 | |
| WUDI-MergingBackbone=ViT-B-32, Setting=FFT2026.03 | 69.9 | 76.71 | |
| ZeroshotBackbone=ViT-B-32, Setting=FFT2026.03 | 56.1 | — |