can the medium model run on CPU
I can get the large model to run on the cup but just not the medium for some reason. My best guess is that it's because of the different architectures between the models and the medium's reliance on xformers? To be clear, both models run perfectly fine using "cuda" and the large works using "cpu," it's just the medium I can't get to work on the "cpu."
Anyways, here's the error I get just for the medium model:
Apologies for the delay in getting to your issue. Can you please provide the details of how you're running the model? Providing a code snippet that reproduces the issue would be a great help.
Also, are you possibly trying to enable the fast xformers attention GPU kernel on CPU? I will admit I haven't tested this architecture on CPU at all, but I am fairly confident the xformers kernel is GPU only and must be disabled on CPU.