What are the dependencies for a Ray worker?

How severe does this issue affect your experience of using Ray?

  • None

I need to use a different version of CUDA than the one provided in the Ray ML docker images. This requires me to build my own docker image with Ray installed. What are the Python/system dependencies I need for a Ray worker node?

An alternative to this, would be if I could specify a CUDA version for the ray gpu image.

Edit: I see that I can choose the CUDA version. While I have a slight desire to know what goes into my Docker images, that’s a problem for another day.

Generally speaking, you just need ray cli to be on your $PATH and the python package to be on your $PYTHONPATH by default (e.g. conda env need to be the default).

It sounds like you may also be interested in poking around the docker files where this is defined. ray/docker at master · ray-project/ray · GitHub

Depending on the environment you’re deploying in you may also need some cloud dependencies.