pretrained Deberta models fail to load in 3.1

[sanderland]

Environment info

• adapter-transformers version: master / 3.1a0

Information

Model I am using: Deberta

When loading it using model = AutoAdapterModel.from_pretrained("microsoft/deberta-base")

It returns a size mismatch for deberta.encoder.layer.[n].attention.self.in_proj.weight: copying a param with shape torch.Size([2304, 768]) from checkpoint, the shape in current model is torch.Size([768, 2304]). Same for "HannahRoseKirk/Hatemoji"

Additional tests

• deberta-v3 (=V2 in implementation) works
• adapter-transformers 3.0.1 does not have deberta