pretained "vit_deit_base_distilled_patch16_384" with custom "num_classes"

[rajaboja]

I tried this,

timm.create_model("vit_deit_base_distilled_patch16_384",pretrained=True, in_chans=1,num_classes=11)

it raises,

RuntimeError: Error(s) in loading state_dict for DistilledVisionTransformer:
	size mismatch for head_dist.weight: copying a param with shape torch.Size([1000, 768]) from checkpoint, the shape in current model is torch.Size([11, 768]).
	size mismatch for head_dist.bias: copying a param with shape torch.Size([1000]) from checkpoint, the shape in current model is torch.Size([11]).

This is only happening for distilled models. Is this expected?
Can someone please help?

[rwightman]

@rajaboja it's expected with issue #433 and PR #434 .. I have not had a chance to address it and integrate the PR with some additional changes i have to those models

[rajaboja]

Many thanks! @rwightman