On GCP cuda is available pre fastai install, but not after

I bootup a GCP instance as per instructions.

I run python, import torch and print(torch.cuda.is_available()) is True.

However, after I run conda install -c fastai fastai, I see python-cpu being installed. Once the conda install finishes, print(torch.cuda.is_available()) produces False.

This is not remedied by a susquent conda install -c fastai pytorch-gpu

Any thoughts on what may be going on?

I installed fastai v1 instead as suggested elsewhere in the docs and all works! It may be valuable to point this out on the Google Cloud page.