If you run Ray_on_AML with python 3.10 or newer, all jobs in AML will fail because the environment creation fails. This is probably due to the create environment function inheriting the python version from the user (and not allowing it to be changed through kwargs) and conda not supporting python 3.10 without updating some libraries beforehand.
Expected behaviour:
- Either being able to specify python version of environment in case the python version we want running on the cloud is different than the user's.
- or Not relying on a conda environment and just a python docker image with a requirements file.