Pytorch includes the necessary ROCm files for image gen as far as I know. My computer does have an /opt/ROCm folder but it’s only a few megabytes in size.
Unfortunately I don’t see any mention of pytorch in your linked program, and it does mention a proper ROCm install as a requirement there.
There are different pytorch install commands for cuda/ROCm. Usually the key to getting ROCm to work with AI GitHub projects is to edit a config file/launch option and replace the pytorch command with a ROCm one