Raimondas Galvelis

@risi-kondor thanks for bringing `GELib` to out attention. Do you have a development roadmap for the library? Would you be interested in a collaboration with NNPOps developers?

Yes, this should be easy to implement.

@davkovacs I'm working on this (https://github.com/openmm/NNPOps/pull/61).

Answering to https://github.com/peastman/NNPOps/pull/5#issuecomment-706441385 > Looks like the neural net part is now the bottleneck. From the benchmarks in #6, doing both forward and backward passes through the features for a...

@peastman, just letting know before you start writing the NN part in CUDA directly. I almost have a working implementation of the NN part using the batched matrix multiplications. I...

@isayev In case of ANI-2x, for small molecules (~100 atoms), the bottleneck is the matrix multiplications in the dense layers. So, a single-model NNP (rather than the ensemble) would improve...

Because I didn't need PBC, so I cut some corners.

I think that we choose one of the existing formatting styles and adapt it (even if it needs reformating the existing code). For Python, it could be `black`. For C++,...

I have purged the GA cache. If it fails, try to rerun.

``` [Linux (CUDA 11.8, Python 3.10, PyTorch 2.0)](https://github.com/openmm/NNPOps/actions/runs/5892449251/job/15981745203#step:1:39) You are running out of disk space. The runner will stop working when the machine runs out of disk space. Free space...