The universal-training collection fails to build on the s390x architecture. The build fails because Triton is pulled in as a transitive dependency, and Triton does not support the s390x platform. 

    1.Analyze why Triton is pulled into the collection
    2. Try to come up with a workaround

look at the dependency graph file. You can find it under mnt/work-dir/graph.json in the job artifacts of the failed job. fromager graph why graph.json triton will tell you what packages depend on Triton.

    https://gitlab.com/redhat/rhel-ai/rhai/pipeline/-/jobs/12750904937