Cannot Load Models - Missing VisualTransformer in open_clip

[aryajalali]

Hi. Nice paper.

When trying to load a model checkpoint using torch.load(), the following error occurs:

AttributeError: Can't get attribute 'VisualTransformer' on <module 'open_clip.model' from 'C:\\Python311\\Lib\\site-packages\\open_clip\\model.py'>

It seems that open_clip no longer contains VisualTransformer, but the checkpoint expects it.

[EnnengYang]

Hi,

This is probably due to the different versions of open_clip we are using. In my development environment, open-clip-torch==2.0.2, torch==1.8.1+cu111, you can try this open_clip version.

References:
EnnengYang/RepresentationSurgery#1

Thanks!