The opendatahub-io/notebooks repository provides a collection of container images and notebook environments designed for data science, machine learning, and research workflows within the OpenDataHub ecosystem. These workbench images are built to integrate with the ODH Notebook Controller and are published to quay.io/repository/opendatahub/workbench-images for deployment in Kubernetes environments.
The repository maintains a diverse portfolio of notebook images supporting multiple hardware configurations and frameworks. The image inventory includes CPU-based workbenches with minimal and data science configurations, GPU-accelerated variants using CUDA 12.8 with PyTorch and TensorFlow support, and AMD ROCM 6.3 alternatives. All images are built on UBI9 with Python 3.12 and support multiple architectures including x86_64, aarch64, ppc64le, and s390x, though GPU variants have more limited architecture support. Specialized images include a TrustyAI workbench for CPU environments and CodeServer options for IDE-based development.
The build infrastructure uses a hermetic approach with Konflux and Cachi2 for reproducible builds. The repository employs a shared prefetch-input directory at the repository root containing RPM lock inputs and dependency specifications, with symlinks from individual image directories. Lock files are managed through uv with a dual versioning policy: development uses a flexible uv version range while image builds use a stricter pinned version for consistency. Base image versions are centrally configured through a versions_config.yml file that drives synchronization of build arguments and RHDS channel resolution.
Development and testing infrastructure relies on Testcontainers.com for running container self-tests from Python, with consideration for both Docker/Podman and Kubernetes execution environments. The repository includes Playwright-based browser testing and pytest-based test suites. Testing can be performed locally or deployed to Kubernetes clusters using make targets like deploy8 and deploy9 for UBI8 and UBI9 variants respectively.
The repository requires Python 3.14, podman or docker, the uv package manager, make, curl, and git with git-lfs support for development. Documentation is provided through a wiki page explaining the workbench architecture, with additional technical documentation covering base image version update configuration, testing procedures, and lock file generation workflows.
