pesmaker train-setup

train-setup prepares model training inputs and a training submit.sh.

Normal users can let next run this stage after the dataset exists.

Use

pesmaker train-setup run.yaml

Minimal YAML

project: train_run

training:
  model: nep
  output_dir: training
  dataset: train.xyz
  command: nep

jobs:
  submit_command: sbatch
  cores_cpu: 36
  sub_file:
    training: templates/sbatch/nep.sh

Outputs

training/
  train.xyz
  nep.in
  training_manifest.jsonl
  submit.sh

For NEP, PESMaker writes a starter nep.in using GPUMD defaults and infers the type line from train.xyz when that file exists. You can override NEP parameters under training.nep:

training:
  model: nep
  output_dir: training
  dataset: train.xyz
  nep:
    cutoff: [8, 4]
    neuron: 30
    generation: 100000

Two-step NEP training

Enable two-step training with training.two_step: true:

training:
  model: nep
  output_dir: training
  dataset: train.xyz
  two_step: true

The first run writes training/step1 with:

lambda_e      0.2
lambda_f      2
lambda_v      0.1

After training/step1/nep.txt exists, run pesmaker next train.yaml again. PESMaker copies the full step1 folder to training/step2, rewrites nep.in, and switches to:

lambda_e      2
lambda_f      1
lambda_v      1

Next Step

Preview or submit training:

pesmaker submit run.yaml --stage training --dry-run
pesmaker submit run.yaml --stage training

With next, PESMaker writes the dry-run log and prints the real submit command.