Closed peastman closed 6 months ago
We really need to fix the CI...
There should also be a Python side test. Take this one if you'd like: https://github.com/openmm/openmm-torch/blob/4c9b5e509d15ce52db7600f7538fbe66538c1c2a/python/tests/TestParameterDerivatives.py
BTW, this is still a problem. I do not think OpenMM-Torch can do anything about it, but perhaps there is a way to detect it and provide a useful message?
Thanks for the comments. It turned out that what I had written didn't work with CUDA graphs. I restructured it to handle the input tensors in a different way and made the test case run both with and without graphs. Can you see if it looks better now?
Is this ok to merge? Getting CI working is its own major project. I'm working on that in another PR.
Implements #141.