Multi-task Image Classification on Vision Tasks Suite (20 tasks) (ViT-B/32, test)
91.4AccuracyIndividual
Evaluation results
| Method | Links | ||
|---|---|---|---|
| IndividualBackbone=ViT-B/322026.06 | 91.4 | — | |
| ReTeXBackbone=ViT-B/322026.06 | 89.8 | 98.3 | |
| MoW-MergingBackbone=ViT-B/322026.06 | 79.3 | 86.8 | |
| WEMoEBackbone=ViT-B/322026.06 | 74.4 | 81.2 | |
| AdaMergingBackbone=ViT-B/322026.06 | 66.2 | 72.7 | |
| Consensus TABackbone=ViT-B/322026.06 | 65.4 | 72 | |
| TIES-MergingBackbone=ViT-B/322026.06 | 62.3 | 68.8 | |
| Weight AveragingBackbone=ViT-B/322026.06 | 61 | 67.5 | |
| Task ArithmeticBackbone=ViT-B/322026.06 | 60.5 | 66.7 | |
| Twin-MergingBackbone=ViT-B/322026.06 | 57.5 | 61.8 | |
| Zero-shotBackbone=ViT-B/322026.06 | 56.1 | — | |
| DaWinBackbone=ViT-B/322026.06 | 52.8 | 57.7 |