Iván Pulido

@richardjgowers thanks for reporting the issue and contributing the possible solution. I tend to agree with your remarks, the Python garbage collector should already be doing this. Nevertheless, I would...

@mkrompiec Thanks for your interest in the tools in our ecosystem! It is indeed related to both of those things. We have been lacking on this front for a while,...

> @ijpulidos also mentioned to me that the energy discrepancy on CPU is not present when using one of the OpenMM nightly dev builds -- Iván, maybe you could provide...

We are not seeing the issues with the `openmm 8.0.0dev2 conda-forge/label/openmm_dev` package. So this might be solved now. That said, we did see some discrepancies between platforms/OS for the `openmm...

> Is there a way to reproduce it? In the comment linked above (linked here again https://github.com/choderalab/perses/pull/1124#issuecomment-1307599051 ) there's a package with a script and serialized system/state to reproduce the...

Closing since I have not seen these issues with the latest dev build.

On a second review of the github actions, we are still seeing the issues [here](https://github.com/choderalab/perses/actions/runs/3566740075/jobs/5993568339#step:9:1625). Reopening, will try to come up with a minimally reproducible example again for this.

I'll contact Openeye support with this and we'll see what they have to say. Thanks for taking the time to dig into this!

@peastman Actually, thinking about this, do you know how this explains that we get the issue with Openmm8 rc and not with current Openmm7.7 packages?

Tests on the perses side are passing now with the beta release from Openeye https://github.com/choderalab/perses/pull/1237