@lexming @smoors Does this ring a bell for you?

The person who requested it mentioned that it's already available on Hydra...

I've asked for some more input from the person who requested this: which installation they're using on Hydra @ VUB, and how they're using this exactly (since it seems like there's a bunch of options when installing it)

Try on top of foss/2023a with TensorFlow + PyTorch + jax as dependencies.

Try first CPU-only, then also with GPU/CUDA if that works.

Start with max. 2 hours to see how far you get.

@lexming @smoors Does this ring a bell for you?

The person who requested it mentioned that it's already available on Hydra...

don't think we have it, the user indicates he was wrong..

More details on requirements:

Considering the backends: DeePMD is most often combined with TensorFlow (tensorflow or
tensorflow-gpu), so I recommend installing it with that option. If that doesn't work,
the second option would be PyTorch although this is less preferred. Dpdata is
optional, but preferred. Compiling with C++ is a must (so it can interphase with
LAMMPS later for validation).

@pavelToman So, give it a try with TensorFlow + PyTorch + jax, and quickly focus on TensorFlow if either PyTorch or jax are problematic for some reason.

There is the same problem as for Horovod when it run pip install deepmd_kit: #390 (comment)

#error This file was generated by an older version of protoc
#error incompatible with your Protocol Buffer headers
#error regenerate this file with a newer version of protoc.

EDIT: continue with the same protobuf hack

For now I have prepared easyconfigs for:

1. DeePDM-kit-3.0.1 - python module + c++ module + cpu version + TensorFlow/Jax version
2. DeePDM-kit-3.0.1 - python module + c++ module + cpu version + TensorFlow/Jax version + LAMMPS (in plugin mode)
3. DeePDM-kit-3.0.1 - python module + c++ module + gpu version (CUDA-12.1.1) + TensorFlow/Jax version
4. DeePDM-kit-3.0.1 - python module + c++ module + gpu version (CUDA-12.1.1) + TensorFlow/Jax version + LAMMPS (in plugin mode)

I was unable to build horovod with PyTorch support, so I am unable to build DeePDM-kit with PyTorch support

PR - CPU version:

• {ai}[foss/2023a] DeePDM-kit v3.0.1 (+ version with LAMMPS plugin) , Horovod v0.28.1 w/ TensorFlow 2.13.0 easybuilders/easybuild-easyconfigs#22217

PR - GPU version:

• {ai}[foss/2023a] DeePDM-kit v3.0.1, Horovod v0.28.1 w/ CUDA 12.1.1 TensorFlow 2.15.1 easybuilders/easybuild-easyconfigs#22219