Kubeflow is an open-source platform for machine learning and MLOps on Kubernetes, introduced by Google. It helps people build, train, and use machine learning models more easily by organizing the steps of machine learning into separate tools.
Kubeflow includes tools for writing code, training models, and using models to make predictions. Some of these tools are Kubeflow Notebooks for writing and running code, Kubeflow Pipelines for organizing the steps of training a model, and KServe for using trained models to make decisions.
One special feature of Kubeflow is that you don’t have to use all of its tools at once. Each part can be used on its own, so you can pick just the tools that fit your project. This makes Kubeflow very flexible for different kinds of machine learning work.
History
The Kubeflow project was first announced in 2017 at KubeCon + CloudNativeCon North America by Google engineers. It began as a way for Google to share how they used TensorFlow inside their company.
The first version of Kubeflow came out in 2018. By March 2020, many of its parts were ready for real use. In 2022, Google asked to join the Cloud Native Computing Foundation with Kubeflow. In 2023, the foundation agreed to include Kubeflow as an incubating project.
Components
Kubeflow has tools to help build and use machine learning models. Kubeflow Notebooks lets users develop models using web-based environments like Jupyter Notebook, Visual Studio Code, and RStudio.
Kubeflow Pipelines helps train models by creating workflows using Docker containers. KServe is used to deploy trained models for different frameworks such as TensorFlow and PyTorch. Katib automates parts of machine learning, like testing many settings to find the best one.
Release timeline
Kubeflow has had many updates since it started. These updates add new tools to help with machine learning. Each update makes using machine learning on Kubernetes easier and stronger. The timeline shows when these updates happened and what new features were added.
Safekipedia