Thanks for sharing! It would be amazing if GitLab would support this for self-hosted instances. MR especially would improve immensely when adding an AI based merge request bot that catches common mistakes, like copy paste errors or uninitialized variables (c++).

@kelteseth Interesting suggestion. I haven't tried but I think this would be possible already and easier than building natively, since Docker images are provided with all necessary software + ROCm installed. You could register a self-managed Runner with Docker Executor on a machine with a 7900XTX, then build ML workloads directly into your .gitlab-ci.yml.

Just be sure to use a docker image with ROCm 5.6+ and use the HSA_OVERRIDE_GFX_VERSION envvar when running workloads.

Oh that seems nice I have a 7900xtx but would like to run some pytorch. This is discussion is the same thing as running it ubuntu 22.04? It's a bit of a pain in the ass trying to boot it up from there instead of having it run normally from VScode. Any recommendations or suggestions in running pytorch / tensorflow models with the 7900xtx would be greatly appreciated. :)