This is open-sourced code and data for paper entitiled "Breaking the Speed, Power, Cost and Size Limits of Molecular Dynamics with Ab Initio Accuracy".
The code
folder, which contains the generation that implements mdpu (Molecular Dynamics Processing Unit) training. You can install and use the code according to the documentation.
For each material, two folders (train
and md
) are kept in its corresponding directory.
The train
contains the input scripts and results of the training. The training step is divided into 2 steps to train CNN and QNN. You can type dp train-mdpu train_cnn.json -s s1
to train CNN and get the corresponding mdpu_cnn
folder. You can type dp train-mdpu train_qnn.json -s s2
to train QNN and get the corresponding mdpu_qnn
folder. The file mdpu_qnn/model.pb
is the model file used to perform the mdpu MD.
The md
contains the input scripts for the mdpu MD, the model files, the output files, and the post-processing files for the results. You can run mdpu MD by uploading input scripts and model files to an online server and running MD through the Bohrium.
For more details about train
and md
, please refer to the documentation.