Kubeflow SDK¶

A Pythonic API to Run AI Workloads at Scale

Run any AI workload at any scale — without the need to learn Kubernetes. The Kubeflow SDK provides simple and consistent Python APIs across the Kubeflow ecosystem, enabling you to focus on building AI applications rather than managing infrastructure.

Quick Start¶

Get up and running in minutes.

pip install kubeflow

from kubeflow.trainer import TrainerClient, CustomTrainer

def train_fn():
    import torch
    model = torch.nn.Linear(10, 1)
    print("Training complete!")

client = TrainerClient()
client.train(trainer=CustomTrainer(func=train_fn, num_nodes=3))

Full installation guide →

Why Kubeflow SDK?¶

🎯 Unified Experience

Single SDK to interact with multiple Kubeflow projects through consistent Python APIs.

🐍 Simplified AI Workloads

Abstract away Kubernetes complexity using familiar Python APIs.

🚀 Built for Scale

From local laptop to production cluster with thousands of GPUs using the same APIs.

💻 Local Development

First-class support for local development without a Kubernetes cluster.

Supported Projects¶

Project	Status	Description
Trainer	✅ Available	Train and fine-tune AI models with various frameworks
Katib	✅ Available	Hyperparameter optimization
Model Registry	✅ Available	Manage model artifacts and versions
Spark	✅ Available	Data processing and feature engineering
Pipelines	🚧 Planned	Build, run, and track AI workflows
Feast	🚧 Planned	Feature store for machine learning

Getting Involved¶

Join the community and help shape the future of ML on Kubernetes.

💬 Community

🤝 Contribute

📚 Resources