nep

Description:
Trains potential functions using NEP.

Input Parameters

Usage:

NepTrain nep [options]

Options:

• -dir, --directory
  Path for NEP calculations. Default: ./cache/nep.

• -in, --in
  Path to nep.in file. Default: ./nep.in.

• -train, --train
  Path to train.xyz file. Default: ./train.xyz.

• -test, --test
  Path to test.xyz file. Default: ./test.xyz.

• -nep, --nep
  Path to potential function file. Default: ./nep.txt.

• -pred, --prediction, --pred Enable prediction mode. Default: False.

• -restart, --restart_file, --restart Path to restart file. Default: None.

• -cs, --continue_step Steps to continue from restart. Default: 10000.

Output

Training results are written to the directory specified by --directory (default ./cache/nep). Important files include:

• nep.txt – trained potential function.

• loss.out – training loss for each step.

• nep.out / nep.err – standard output and error logs.

• nep_result.png – plot of the training loss.

Example

Run NEP training with explicit training and testing datasets:

NepTrain nep --train train.xyz --test test.xyz

Resume training from a restart file for an additional 50,000 steps:

NepTrain nep --restart_file ./cache/nep/nep.restart --continue_step 50000