The training part of MLOps is important to ensure replicability and other desirable properties or the ML artifact, but the rest is clearly good CI/CD and observability.