Kubeflow

Kubeflow is a self-hosted AI platform for teams developing and running machine learning workloads on Kubernetes. Its modular projects cover interactive development, training and the work of moving models into production.

• Development environments: Kubeflow Notebooks provides interactive workspaces for AI, machine learning and data work on Kubernetes.
• Distributed training: Kubeflow Trainer supports model training and LLM fine-tuning with PyTorch, MLX, HuggingFace, DeepSpeed, Megatron and JAX.
• ML workflows: Kubeflow Pipelines lets teams build and deploy portable machine learning workflows, while the Spark Operator manages Spark applications on Kubernetes.
• Automated tuning: Kubeflow Katib searches hyperparameters and neural architectures and supports early stopping during training.
• Model management: Kubeflow Hub organizes models, versions and artifact metadata between experiments and production.

Teams can choose individual components for their own stack. The Central Dashboard brings authenticated web interfaces for Kubeflow and other ecosystem components into one place.

Kubeflow runs on Kubernetes, including local laptop, on-premises and cloud environments, and supports using the same workload code across those deployments. It's open source under Apache 2.0.