The workbench-images repository provides custom container images for use with Open Data Hub and Red Hat OpenShift Data Science. These images are designed to serve as interactive development environments and runtime containers for data science and machine learning workflows on OpenShift platforms. The repository is maintained by Red Hat's AI Services Business Unit and contains the build configurations and tooling necessary to create containerized workbenches with various combinations of IDEs, data science libraries, and machine learning frameworks.
The repository offers both pre-built workbench images and runtime images. Workbench images include integrated development environments such as Jupyter, VSCode, RStudio, Langflow, and StableStudio, allowing users to work interactively within OpenShift Data Science. Runtime images contain the same library bundles as their workbench counterparts but exclude the IDE components, making them suitable for executing data science pipelines. All images are based on CentOS Stream 9 with Python 3.11, except for the Langflow image which uses Python 3.9 due to compatibility constraints.
The images are constructed modularly using a bundle system that allows flexible composition of packages and applications. Available bundles include Minimal, Data Science, PyTorch, Tensorflow, LangChain, Langflow, OptaPy, R, and Spark. The Data Science bundle includes standard tools like pandas, scikit-learn, matplotlib, and Kubernetes client libraries. Specialized bundles add frameworks such as PyTorch 2.0.1, Tensorflow 2.13.0, or LangChain 0.0.285. GPU-accelerated variants with CUDA support are available for PyTorch, Tensorflow, and LangChain bundles. The base OS layer includes numerous pre-installed packages such as FFmpeg, GStreamer, database clients for MongoDB, MSSQL, MySQL, and PostgreSQL, along with Git LFS and other development tools.
For users whose needs are not met by pre-built images, the repository provides an interactive wizard script that guides users through creating custom images. The wizard, invoked via the interactive-image-builder.sh script, requires Podman, Skopeo, and bc to be installed and generates build recipes based on user selections of bundles and IDEs.
According to GitGenius activity tracking, the repository shows a median issue and pull request response latency of 7.6 hours across tracked items, with a mean latency of 174.8 hours. The most active contributors tracked include shalberd with 5 events, followed by guimou and koep with 2 events each. The repository shares contributors with related projects including opendatahub-io/odh-dashboard, elyra-ai/elyra, and opendatahub-io/notebooks, indicating integration within a broader ecosystem of Open Data Hub and data science tooling projects.
