fine_tune

This keyword instructs nep to fine tune a model starting from a foundation model.

Syntax

fine_tune <nep_model_file> <nep_restart_file>

Parameters

<nep_model_file> is the potential file for the foundation model and <nep_restart_file> is the restart file for the foundation model.

Output

None.

Examples

The example below uses the foundation model in potentials/nep/nep89_20250409.

The hyperparameters that define the architecture of the foundation model cannot be changed and must be repeated in nep.in. In this nep.in template, replace type <your types> with the number and chemical symbols of the elements to train:

# for fine-tuning
fine_tune nep89_20250409.txt nep89_20250409.restart
type <your types>

# These cannot be changed:
version    4
zbl        2
cutoff     6 5
n_max      4 4
basis_size 8 8
l_max      4 1 1
neuron     80

# These can be changed:
lambda_1   0
lambda_e   1
lambda_f   1
lambda_v   1
batch      5000
population 50
save_potential 1000 0 1
generation 5000

Notes

None.