Summary

Source: https://github.com/deepqmc/deepqmc

License: MIT License

Path: /software/DeepQMC

Documentation: https://deepqmc.github.io/


DeepQMC implements variational quantum Monte Carlo for electrons in molecules, using deep neural networks written in PyTorch as trial wave functions. Besides the core functionality, it contains implementations of the PauliNet ( https://doi.org/ghcm5p)

Using  DeepQMC

 To initialize the environment (it's actually a python 3.9 virtual environment) use the module command:

module load maxwell
module avail DeepQMC
#  ------ /software/etc/modulefiles -----
#  DeepQMC/0.3.1 DeepQMC/0.3.1c

module load DeepQMC
python -c 'from deepqmc import Molecule'

Installation

The pure pip installation (v0.3.1) is largely ok, but tensorboard is deficient. A conda-based installation (v0.3.1c) looks considerably better. The installation recipe used:

module load maxwell conda
. conda-init
conda create --prefix=/software/DeepQMC/0.3.1c python=3.8
conda activate /software/DeepQMC/0.3.1c
conda install -c pyscf -c conda-forge numpy scipy pytorch toml uncertainties pyscf tensorboard tensorboard-data-server pillow tqdm h5py click tomlkit
pip install -U deepqmc[wf,train,cli]

# needs cudatoolkit=11.3 to support A100
conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch

# "default" h5py is broken. See https://github.com/h5py/h5py/issues/1880
conda install -c conda-forge hdf5=1.12 h5py

Tensorboard

When specifying a run-directory in your code, you can view the progress using tensorboard, e.g.

tensorboard --logdir runs/ --bind_all
TensorFlow installation not found - running with reduced feature set.

NOTE: Using experimental fast data loading logic. To disable, pass
    "--load_fast=false" and report issues on GitHub. More details:
    https://github.com/tensorflow/tensorboard/issues/4784

TensorBoard 2.8.0 at http://max-display008.desy.de:6006/ (Press CTRL+C to quit)

Opening the tensorboard in your browser should roughly look like this:

 

 


Attachments:

deepqmc-1.png (image/png)