How is parallelism determined when installed with pip?

[joshkamm]

I'm curious how xtb-python decides how many threads / processes xTB should use when installed through a simple pip install xtb? Technically I include xtb as a dependency in my setup.py and install my own project with pip.

For context, my lab mate and I are running the same script on the same computing cluster that makes repeated calls to xtb-python, but each energy / gradient calculation runs about an order of magnitude slower for me (a few seconds instead of a few tenths of a second on an 84 atom Pd complex). When I run top on the linux node it looks like his xTB is running on one cpu and mine is trying to use all of the node's cpus. I'm trying to run xTB on many geometries, so it seems like running them simultaneously each using one cpu would be more efficient than an individual xTB calculation using multiple cpus anyway. Since we're running what I believe to be identical python code I think environment variables and such may be contributing.

Thanks for working on a great open source project!