You can create a new conda environment and start the Ray cluster head using the new conda environment. Then, all the workers can connect to the cluster using the new environment as well. Here's an example snippet:

setup: |
  conda activate myray
  if [ $? != "0" ]; then
    conda create -n myray -y python=${PYTHON_VERSION}
    conda activate myray
  fi
  pip install <https://s3-us-west-2.amazonaws.com/ray-wheels/master/${RAY_COMMIT}/ray-3.0.0.dev0-cp38-cp38-manylinux2014_x86_64.whl>

  # works for single node (if you need to do multiple nodes, we can talk more about it)
  ray start --head

run: |
  # you should be able to connect to your own ray cluster in your program
  conda activate myray
  python myprogram.py