Error with torch.hub.load when using num_classes kwarg

[kchonka]

When trying to access dpn107 via the PyTorch Hub API, I run into this error:

Error(s) in loading state_dict for DPN: size mismatch for classifier.weight: copying a param with shape torch.Size([1000, 2688, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 2688, 1, 1]). size mismatch for classifier.bias: copying a param with shape torch.Size([1000]) from checkpoint, the shape in current model is torch.Size([16]).

The call I used was:
torch.hub.load('rwightman/pytorch-dpn-pretrained', 'dpn107', num_classes=16, pretrained=True)

When specifying the num_classes arg, this error appears.