While the Ray base images (especially the ray-ml flavor) come with a comprehensive set of packages frequently used for machine learning, you may still need to modify your environment when you need additional packages or when you need a specific version of a given package.
Domino allows you to easily package and manage dependencies as part of your Ray-enabled compute environments. This approach creates the flexibility to manage dependencies for individual projects or workloads without having to deal with the complexity of a shared cluster.
To add a new dependency, you must add the appropriate statements in the Docker Instructions section of the relevant Ray base and Ray execution compute environments.
For example, if you wanted to add a particular version of PyTorch you might include the following.
### Optionally specify version if desired
RUN pip install torch==1.8.0 torchvision==0.9.0