tesla

TESLA Workflow

A bash script based workflow for training a machine learning potential automatically.

Introduction

Train-Explore-Screen-Label Active-learning (TESLA) workflow is a bash script based workflow for training a machine learning potential automatically.

This workflow is inspired by dpgen and ai2-kit.

It is a bash script built with oh-my-batch and ai2-kit, which makes it easy to customize. Developers can easily add their own steps to the workflow by modifying the bash script directly.

Getting Started

To run the workflow, you need to ensure your environment has Python 3. And then all you need to do is to run the following command:

./run.sh

To customize the workflow, you can:

• Modify configuration in 00-config folder, which include template file of DeepMD, LAMMPS, CP2K, Slurm, etc.
• Modify training strategy by creating your own scripted by copying 01-workflow/iter-basic-dp-lammps-cp2k.sh and modify it.
• Add more iterations by editing run.sh.

Send PR if you have any good ideas to improve the workflow.