We'll need to make the necessary changes in https://gitlab.com/redhat/rhel-ai/rhaiis/containers to support versioned variants.

The variant names will have to change:

Example for CUDA:

• Current: `cuda-ubi9`
• New: `cuda-12.8-ubi9`, `cuda-13.0-ubi9`, etc.
• Format: `cuda-{major}.{minor}-{base-os}

We need to work hand to hand with the Inference team and also the AIPCC Productization team to ensure we don't break the pipelines and test every change we make.

The initial target of this work is to support both CUDA 12.8 and 12.9, where the model-opt image would need to be on 12.9. This is the thread for that discussion.