End-to-end example of training a model on a multi-node multi-GPU Kubernetes cluster in GKE using the Shared mode that allows consistent logging to the same run ID from multiple independent processes.