An issue has been identified where on machines using the fast.ai linux template, running torch.cuda.is_available() returns False.
We were able to resolve this issue by updating the Nvidia driver with:
sudo apt update && sudo apt upgrade
When you are prompted to update minion/salt, please choose N.
New machines created as of June 18, 2018, should not encounter this issue, as the template was fixed and updated. If the above does not resolve the error for you, please open a support ticket and include your machine's hostname and, if possible, a screenshot.