[FEATURE] Improve job scheduling when a function uses pytorch

[badmonster0]

What is the use case?
Currently we're using a global mutex to make one function (using pytorch) is executed at a time, to avoid conflicts. We should improve the scheduling strategy.

Describe the solution you'd like
We may try to use multiprocessing for this.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.