Great paper and cool toolkit for implementing different versions of PET!
Did you consider updating your code a bit to make it into a PYPI package / or modularising the API more? At the moment users have to clone your repo and can mostly rerun your experiments. I'd be very interested in running different methods on my own data in a google colab notebook and it would be very useful if I could pip install the code and then decide to, e.g. only import an ADAPET class and then run a fewshot_trainer class on my own data integrated in my own custom colab notebook script. I understand that this would require additional work, but I think many more people would start using your code.