self-supervised and fastai=2.7.5 incompatibility

[hamish-haggerty]

Hello,

It seems the tutorials don't work with the current version of fastai / pytorch. After installing self-supervised the version of fastai installed is 2.7.5 and torch is 1.11.0.

After downloading the Barlow Twins tutorial https://colab.research.google.com/github/KeremTurgutlu/self_supervised/blob/master/nbs/14%20-%20barlow_twins.ipynb all the cells run ok up to:

fastai_encoder = create_encoder('xresnet18', n_in=1, pretrained=False)

which throws the error,

AttributeError: 'function' object has no attribute 'named_parameters'

I downgraded fastai and pytorch to the versions that exists in colab and then the tutorial works fine. However there are some bugs in old versions of fastai that are limiting what I am trying to do with self-supervised.

Is self-supervised simply incompatible with new versions of fastai?

Thank you for any help at all.