Multi-task Image Classification on Vision Tasks Suite (50 tasks) ViT-B/32 (test)
90.8AccuracyIndividual
Evaluation results
| Method | Links | ||
|---|---|---|---|
| IndividualBackbone=ViT-B/322026.06 | 90.8 | — | |
| ReTeXBackbone=ViT-B/322026.06 | 81.5 | 89.8 | |
| Consensus TABackbone=ViT-B/322026.06 | 52.1 | 57.4 | |
| WEMoEBackbone=ViT-B/322026.06 | 51.6 | 56.8 | |
| AdaMergingBackbone=ViT-B/322026.06 | 49.4 | 54.4 | |
| Weight AveragingBackbone=ViT-B/322026.06 | 47.8 | 52.6 | |
| TIES-MergingBackbone=ViT-B/322026.06 | 47.5 | 52.3 | |
| Zero-shotBackbone=ViT-B/322026.06 | 46.8 | — | |
| Task ArithmeticBackbone=ViT-B/322026.06 | 46.8 | 51.5 | |
| Twin-MergingBackbone=ViT-B/322026.06 | 36.4 | 40.1 | |
| DaWinBackbone=ViT-B/322026.06 | 33.1 | 36.5 | |
| MoW-MergingBackbone=ViT-B/322026.06 | 30.1 | 33.1 |