Anyscale base images
Anyscale provides a set of pre-built base container images to use with workspaces, jobs, and services. These base images extend the rayproject/ray and rayproject/ray-ml Docker images.
For each supported version of Ray, Anyscale releases base images with pre-configured Python and Debian packages. Anyscale tests these environments to ensure the installed packages are compatible with the version of Ray, Python, and CUDA installed on the cluster.
These environments include the packages required by most common Ray workloads. You can modify these environments with new libraries and use Anyscale tools to manage these dependencies as you move your applications from development to production.
Image naming conventions
Anyscale uses the following naming convention for base images:
anyscale/ray:<ray-version>-<image-type>-<python-version>-<cuda-version>
For example, the anyscale/ray:2.30.0-slim-py310-cu123 image means the following:
- Ray version is 2.30.0.
- Image type is slim.
- Python version is 3.10.
- CUDA version is 12.3.
Slim images contain the minimum set of dependencies required to run on Anyscale. Anyscale recommends starting with slim images to reduce the size of images, which reduces latency when building images, interacting with image registries, and deploying containers.
Image list by Ray version
See the following pages to view the included packages for each image and version of Ray and Python: