Kubeflow ML on Kubernetes

Kubeflow is an open source machine learning platform designed to make deployments of machine learning workflows on Kubernetes simple, portable, and scalable. The project originated as an internal tool at Google for running TensorFlow on Kubernetes and was released as Kubeflow in December 2017. It is now maintained by the Kubeflow community under the Linux Foundation with contributors from Google, IBM, Cisco, Red Hat, NVIDIA, and others. The platform consists of several composable components that can be deployed individually or together. Kubeflow Notebooks provides Jupyter notebook environments with GPU support directly on Kubernetes clusters with configurable resource quotas and namespace isolation. Kubeflow Pipelines enables building reusable end to end ML workflows using the Kubeflow Pipelines SDK in Python, with features for versioning, artifact tracking, and DAG based execution visualization. KServe provides Kubernetes native model serving with autoscaling including scale to zero, canary rollouts, and support for frameworks including TensorFlow, PyTorch, ONNX, scikit-learn, XGBoost, and HuggingFace. Katib offers hyperparameter tuning and neural architecture search with algorithms including Bayesian optimization, random search, grid search, and Hyperband. The Training Operator supports distributed training for TensorFlow, PyTorch, MXNet, MPI, and PaddlePaddle with native Kubernetes job management. Additional components include multi tenancy management, fairing for building and deploying models from notebooks, and training with Intel optimizations.