dpgen-snakemake

A lightweight Snakemake-based workflow that implements the DP-GEN scheme.

Project structure

• project: Actual project files.
  • workflows: Snakemake workflow files.
  • scripts: Scripts for transforming input / output.
  • templates: Templates for generating files.
  • tests: Tests for the Python scripts.
• project_sidecar: Project files on the CPU machine
• envs: conda environments for workflow orchestration and DeePMD-kit training / evaluation.

Usage

DP-GEN Iteration

In each iteration, specify the environment variable DPGEN_ITER_ID then execute one one of the following rules to generate the corresponding files.

Machine	Name of the rule	Description
GPU	dp_run_train_target	Prepare files for the training step.
GPU	dp_run_model_devi_target	Collect the newly generated potential files, then prepare the input files for the LAMMPS model deviation step.
GPU	dp_run_fp_target	Prepare the structures for the VASP calculation and prepare a tarball for transfer.
CPU	dp_run_fp_input	Collect the tarball of structures from the GPU machine.
CPU	dp_run_fp_target	Collect the structures from the GPU machine, then prepare the input files for the VASP calculation.
CPU	dp_run_fp_done_target	Collect the VASP results and prepare for transfer.
GPU	dp_run_fp_collect2	Collect VASP results from the CPU machine, then prepare the training data for the next iteration.