Commonsense Reasoning on Rainbow large (val)
92.5aNLICompassMTL w/ Tailor
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CompassMTL w/ TailorArch.=Enc only, Tasks=14, Params.=567M, multi-task training=subset of 14 related datasets2022.10 | 92.5 | 88.8 | 96.1 | 88.3 | 82.2 | 90.5 | 89.7 | |
| CompassMTLArch.=Enc only, Tasks=40, Params.=567M2022.10 | 91.7 | 87.8 | 95.6 | 87.3 | 81.7 | 89.6 | 89 | |
| ExDeBERTaArch.=Enc only, Tasks=40, Params.=567M2022.10 | 87.9 | 85.3 | 83.6 | 85.5 | 79.6 | 87 | 84.8 | |
| ExT5Arch.=Enc-Dec, Tasks=107, Params.=770M2022.10 | 82.3 | 85.9 | 89 | 85 | 79.7 | 82.5 | 84.1 | |
| UNICORNArch.=Enc-Dec, Tasks=6, Params.=770M2022.10 | 79.5 | 83.2 | 83 | 82.2 | 75.5 | 78.7 | 80.4 |