Hacker News Reader

Can someone explain me why one would need a container orchestration system to train ml models? Why wouldn't one just stick with containers?

For the same reason you would use a container orchestration system for running anything: easier deployments, scaling, reliability, etc.