Multi-task Image Classification on Vision Tasks Suite 14 tasks (ViT-B/32, test)
90.9AccuracyIndividual
Evaluation results
| Method | Links | ||
|---|---|---|---|
| IndividualBackbone=ViT-B/322026.06 | 90.9 | — | |
| ReTeXBackbone=ViT-B/322026.06 | 89.8 | 98.8 | |
| MoW-MergingBackbone=ViT-B/322026.06 | 83.2 | 91.5 | |
| WEMoEBackbone=ViT-B/322026.06 | 83.1 | 90.7 | |
| DaWinBackbone=ViT-B/322026.06 | 73.8 | 80.3 | |
| AdaMergingBackbone=ViT-B/322026.06 | 70.5 | 77.3 | |
| Consensus TABackbone=ViT-B/322026.06 | 70.4 | 77.4 | |
| Twin-MergingBackbone=ViT-B/322026.06 | 70 | 76.7 | |
| Task ArithmeticBackbone=ViT-B/322026.06 | 65.3 | 72 | |
| TIES-MergingBackbone=ViT-B/322026.06 | 65.1 | 71.9 | |
| Weight AveragingBackbone=ViT-B/322026.06 | 64.3 | 71.1 | |
| Zero-shotBackbone=ViT-B/322026.06 | 57.2 | — |