Consistency of Local and Global Flatness for Federated Learning | ResearchPod