Cannot re-initialize CUDA in forked subprocess

[mkzia]

I get the following error when a qcluster worker attempts to use CUDA:

Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method

How I can get around this error?

[mkzia]

Celery has the ability to share variables between workers. This might be a work around. But does qcluster have global variables?
https://stackoverflow.com/a/70618882