[ANNOUNCEMENT] Kubeflow SDK 0.1.0 is released!

[Andrey Velichkevich]

Hi Kubeflow Community,

We’re excited to announce the very first release of Kubeflow SDK 0.1.0 🎉

pip install -U kubeflow

With just two lines of code, you can now fine-tune LLMs:

from kubeflow.trainer import TrainerClient

TrainerClient().train(
    runtime=TrainerClient().get_runtime("torchtune-qwen2.5-1.5b")

)

We started this project to streamline the AI practitioner experience by providing a Python-native interface to Kubeflow APIs, while hiding Kubernetes complexity under the hood.

With the Kubeflow SDK, you can scale AI workloads easily - without worrying about infrastructure details.

Key Highlights 🚀

✅ Kubeflow Trainer SDK with CustomTrainer and BuiltinTrainers

✅ Local process backend support for local execution of PyTorch training jobs.

✅ TorchTune configuration with BuiltinTrainer.

Learn more:

📣 Kubeflow SDK 0.1.0 release notes: https://github.com/kubeflow/sdk/releases/tag/0.1.0
📣 Kubeflow SDK community calls: https://bit.ly/kf-ml-experience

📣 Kubeflow SDK ROADMAP: https://github.com/kubeflow/sdk/blob/main/ROADMAP.md

A huge thank you to the Kubeflow maintainers who made this release possible and continue to drive the project forward.

Stay tuned for the upcoming features!

Regards,

Andrey