torch-m3gnet¶

🚧This repository is heavily under construction!🚧

This is an unofficial PyTorch implementation of M3GNet. The authors’ reimplementation of MEGNet and M3GNet with PyTorch/DGL will be available here.

Documentation https://lan496.github.io/torch-m3gnet/

Limitations¶

Only support structures with PBC
Only support potential prediction
No support: state, MaterialGraph.phi

Installation¶

See docs/installation.md.

Preparing benchmark datasets¶

See docs/benchmark.md

Development¶

Docuement

sphinx-autobuild --host 0.0.0.0 --port 8000 docs docs_build

GPU container

docker run --gpus '"device=1"' -it -v $(pwd):/app -p 6006:6006 m3gnet

# tensorboard
tensorboard --logdir results/ --host 0.0.0.0 --port 6006

References¶

M3gNet https://github.com/materialsvirtuallab/m3gnet
NequIP https://github.com/mir-group/nequip
MatDeepLearn https://github.com/vxfung/MatDeepLearn
change_transfer_nnp https://github.com/pfnet-research/charge_transfer_nnp
nn-template https://github.com/grok-ai/nn-template

Installation

torch-m3gnet 0.1.dev106+g5d238b3