Since rocm only provides a rhel install method it would be appreciated a silverblue friendly way of using it.
Hello. Unfortunately, we don’t have such container images. Would you want to help us create them? Could you describe what such container should contain?
I was thinking of a toolbox compatible image with features similar of the rocm image provided by docker (and/or the standard tensorflow image). Currently, there is no clear way of doing machine learning on fedora, especially for amd users.
I would like to have a machine learning toolbox / container as well, but I guess that we would need a more up to date rocm version on the host first. Atm. the rocm version available in Fedora is 2.0, where rocm 3.3 is already available.
The whole point of machine learning containers is that you don’t need rocm on your host.
I thought as well, but strangely the official rocm containers from AMD state that the host has to have the latest rocm installed,…