karmada

MLSysOps Open-source Framework

This guidance shows how to deploy a federated Kubernetes environment in Amazon Web Services (AWS) cloud using Amazon Elastic Kubernetes Service (Amazon EKS) and the open source CNCF Karmada project - multi-cluster Kubernetes management system with advanced scheduling capabilities that enable running containerized applications across clusters

Run karmada on mac native

A Python module for modeling Karmada based Multi-Cluster Infrastructures using PetriNets

Demo repository for Karmada showcase and conference talk.

Active Federated Learning: Target-Environment Probes, Active Weight aggregation, and Active Data (BC) fine-tuning. Demonstrated with PPO on CartPole, run locally or on Kubernetes via Kubeflow Pipelines, Temporal-orchestrated workers, and Karmada multi-cluster.

karmada-multi-cluster-multi-ingress setup using karmada

Shared Bash library for bootstrapping local Kubernetes dev clusters with kind, Kubeflow Pipelines, MLflow, MinIO, Temporal, and Karmada. Components deploy via manifests or Juju charms. Consumed as a git submodule, tested with bats and shellcheck.

A Federated Learning framework for distributed Digital Twins on Kubernetes. Simulates diverse robotic environments to collaboratively train a robust RL policy using PyTorch, Temporal, and Kubeflow Pipelines, with Karmada for multi-cluster federation and MLflow experiment tracking.

Built a centralized multi-cluster Kubernetes environment with Karmada, enabling automatic propagation of Deployments, Services, Ingresses, ConfigMaps, and Secrets across two clusters with HAProxy for high availability.