PFD-kit documentation (0.6.2)

PFD-kit is a cloud-based workflow automating the generation of deep-learning based force fields from pre-trained atomic models (P) through fine-tuning (F) and distillation (D) for large scale atomic simulation of practical materials. By exploiting the transferable knowledge in the pre-trained model, force field generation using PFD workflow requires much less training data, saving significant time and computational resources. This feature makes PFD workflow a powerful tool in computational materials science, addressing challenges in training force fields for complex material systems (e.g., high-entropy alloys, surfaces/interfaces) that are previously intractable.

This documentation gives an in-depth view of PFD-kit, which includes background introduction, hand-on tutorials and detailed explanation for input parameters.

The Basics

• Overview
  • Backgrounds
  • Workflows
• Quickstart
  • Easy Install
  • Job submission
  • Tutorial: c-Si

User guide

• Command guide
  • Submitting a Workflow
  • Checking status
  • Restarting from a Checkpoint
  • Downloading Results
  • PFD-kit Command Arguments
• Input script guide
  • Basics
  • Fine-tune task
  • Distillation task
• Arguments of the input script
  • Host and nodes
  • Task definition
  • Structure generation
  • Model training
  • Inference & labeling
  • Exploration
• Examples
  • BNBZTO fine-tune

Developer guide

• API documentation
  • pfd package