Auto-scaling for ML workloads, integrated with the TF workflow - very cool!