Datasets for distillation of CHGNet to DeePMD for α, β, γ phases of Ag₂S with VASP fine-tuning

This repository contains ZIP archives with datasets used for knowledge distillation from a CHGNet model to a faster DeePMD model while preserving accuracy for the three polymorphs (α, β, γ) of silver sulfide (Ag₂S). The datasets include CHGNet data and additional DFT (VASP) reference calculations for fine-tuning. Purpose:These datasets enable the training of a computationally efficient DeePMD potential that matches the accuracy of the more expensive CHGNet model, with additional refinement using high-quality DFT data. Methods: Initial sampling: Molecular Dynamics using CHGNet (Crystal Hamiltonian Graph Neural Network) Reference data: DFT calculations performed with VASP (Vienna Ab initio Simulation Package) Target: Distillation to DeePMD (Deep Potential Molecular DynamicsUsage notes:The data is structured for direct use with DeePMD-kit (*.raw and *.npy formats). Also the training scripts (fine-tune.zip/input.json and fine-tune.zip/finetune.json) are provided. Please arrange *.raw and *.npy data in accordange with input.json and finetune.json files.