Not able to load XCIT (Debendetti2022light)
lukasbm opened this issue · 2 comments
lukasbm commented
I was not able to load the XCIT models.
Currently their entry in the cifar10.py
, cifar100.py
and imagenet.py
files look like this:
('Debenedetti2022Light_XCiT-S12', {
'model': (lambda: timm.create_model(
'debenedetti2020light_xcit_s_imagenet_linf', pretrained=True)),
'gdrive_id':
None
})
When running normally using robustbench.load_model
i get a KeyError as the gdrive_id is not set.
However, even when just running the lambda function timm.create_model( 'debenedetti2020light_xcit_s_imagenet_linf', pretrained=True)
in isolation i get the error that the model is not available on timm.
Can this be fixed by including the gdrive ID's from the original repo?
lukasbm commented
My bad i apparently had robustbench only installed as a third party dependency, and thus had an old version
huanranchen commented
This can be solved by #163