Output files

A training run writes these files to output_dir. predict_dataset writes the prediction files (*_train.out, descriptor.out) in the same format.

Models and checkpoints

File	Contents
nep_best.txt	Best model (lowest validation loss, or training loss without a validation set). The end-of-training prediction uses it.
nep_final.txt	Model at the last epoch.
nep_average.txt	Averaged model; only with use_swa=True.
checkpoint.pt	Full training state, for resuming.
checkpoint_stage1.pt	Full state at the end of stage 1.

All nep*.txt files are GPUMD nep4 / nep4_zbl models.

Logs

File	Contents
output.log	The full console log, including the final E/F/V RMSE / MAE tables.
loss.out	One row per epoch: epoch, loss, RMSE of energy (eV/atom), force (eV/Å), virial (eV/atom) and stress (GPa). With a validation set, four more columns hold the validation RMSEs and the loss column is the validation loss.

Predictions

Written at the end of training with nep_best.txt, every prediction_interval epochs with the current weights, and by predict_dataset. Each row holds the prediction first and the reference second; missing references are written as GPUMD does.

File	Rows	Columns
energy_train.out	one per frame	predicted, reference energy (eV/atom)
force_train.out	one per atom	predicted Fx Fy Fz, reference Fx Fy Fz (eV/Å)
virial_train.out	one per frame	predicted, reference xx yy zz xy yz zx (eV/atom)
stress_train.out	one per frame	predicted, reference xx yy zz xy yz zx (GPa)
*_test.out		the same four files for the validation set
descriptor.out	per frame or per atom	scaled descriptors (predict_dataset(..., output_descriptor=1 or 2))