2_training/forces/train.yaml

# NequIP training config for diamond ML interatomic potential
# Trained on 50 displaced 2x2x2 FCC diamond supercells (QE PBE, ecutwfc=60 Ry)

run: [train, test]

cutoff_radius: 5.0          # Angstrom — covers ~5 neighbor shells in diamond
num_layers: 4
l_max: 2                    # l=2 for better angular accuracy (diamond has sp3 bonding)
num_features: 32            # 32 is safe for 6GB GPU with l_max=2

model_type_names: [C]
chemical_species: ${model_type_names}
monitored_metric: val0_epoch/weighted_sum

data:
  _target_: nequip.data.datamodule.ASEDataModule
  seed: 42

  split_dataset:
    file_path: ${hydra:runtime.cwd}/dataset.xyz
    train: 0.8      # 40 train
    val: 0.1        # 5 val
    test: 0.1       # 5 test

  transforms:
    - _target_: nequip.data.transforms.ChemicalSpeciesToAtomTypeMapper
      model_type_names: ${model_type_names}
    - _target_: nequip.data.transforms.NeighborListTransform
      r_max: ${cutoff_radius}

  train_dataloader:
    _target_: torch.utils.data.DataLoader
    batch_size: 2
    num_workers: 4
    shuffle: true
  val_dataloader:
    _target_: torch.utils.data.DataLoader
    batch_size: 10
    num_workers: ${data.train_dataloader.num_workers}
  test_dataloader: ${data.val_dataloader}

  stats_manager:
    _target_: nequip.data.CommonDataStatisticsManager
    dataloader_kwargs:
      batch_size: 10
    type_names: ${model_type_names}

trainer:
  _target_: lightning.Trainer
  accelerator: gpu
  enable_checkpointing: true
  max_epochs: 2000
  log_every_n_steps: 10

  logger:
    _target_: lightning.pytorch.loggers.CSVLogger
    save_dir: ${hydra:runtime.output_dir}
    name: nequip_diamond

  callbacks:
    - _target_: lightning.pytorch.callbacks.EarlyStopping
      monitor: ${monitored_metric}
      min_delta: 1e-4
      patience: 50

    - _target_: lightning.pytorch.callbacks.ModelCheckpoint
      monitor: ${monitored_metric}
      dirpath: ${hydra:runtime.output_dir}
      filename: best
      save_last: true

    - _target_: lightning.pytorch.callbacks.LearningRateMonitor
      logging_interval: epoch

training_module:
  _target_: nequip.train.EMALightningModule
  ema_decay: 0.999

  loss:
    _target_: nequip.train.EnergyForceLoss
    per_atom_energy: true
    coeffs:
      total_energy: 1.0
      forces: 1.0

  val_metrics:
    _target_: nequip.train.EnergyForceMetrics
    coeffs:
      total_energy_mae: 1.0
      forces_mae: 1.0
  train_metrics: ${training_module.val_metrics}
  test_metrics: ${training_module.val_metrics}

  optimizer:
    _target_: torch.optim.Adam
    lr: 0.005

  lr_scheduler:
    scheduler:
      _target_: torch.optim.lr_scheduler.ReduceLROnPlateau
      factor: 0.5
      patience: 20
      threshold: 0.01
      min_lr: 1e-6
    monitor: ${monitored_metric}
    interval: epoch
    frequency: 1

  model:
    _target_: nequip.model.NequIPGNNModel
    seed: 42
    model_dtype: float32
    type_names: ${model_type_names}
    r_max: ${cutoff_radius}

    num_bessels: 8
    bessel_trainable: false
    polynomial_cutoff_p: 6

    num_layers: ${num_layers}
    l_max: ${l_max}
    parity: true
    num_features: ${num_features}

    radial_mlp_depth: 2
    radial_mlp_width: 64

    avg_num_neighbors: ${training_data_stats:num_neighbors_mean}
    per_type_energy_scales: ${training_data_stats:per_type_forces_rms}
    per_type_energy_shifts: ${training_data_stats:per_atom_energy_mean}
    per_type_energy_scales_trainable: false
    per_type_energy_shifts_trainable: false