model:
    chemical_species: 'univ'  # Ready for 119 elements
    cutoff: 6.0
    channel: 128
    is_parity: False
    lmax: 3
    num_convolution_layer: 3
    irreps_manual:
        - "128x0e"
        - "128x0e+64x1e+32x2e+16x3e"
        - "128x0e+64x1e+32x2e+16x3e"
        - "128x0e"

    weight_nn_hidden_neurons: [64, 64]
    radial_basis:
        radial_basis_name: 'bessel'
        bessel_basis_num: 8
    cutoff_function:
        cutoff_function_name: 'XPLOR'
        cutoff_on: 5.5

    act_gate: {'e': 'silu', 'o': 'tanh'}
    act_scalar: {'e': 'silu', 'o': 'tanh'}

    conv_denominator: 'avg_num_neigh'
    train_shift_scale: True
    train_denominator: False
    self_connection_type: 'linear'

    # Following are used to specify which part of the model would utilize fidelity-dependent parameters for multi-fidelity training.
    # For detailed architecture, please refer to https://arxiv.org/abs/2409.07947
    # Parts using fidelity-dependent weights are indicated as `Modified linear` layers in Figure 1.
    use_modal_node_embedding: False    # If true, use modified linear layer in atom-type embedding layer.
    use_modal_self_inter_intro: True   # If true, use modified linear layers in self-interaction block before the convolution in the interaction blocks.
    use_modal_self_inter_outro: True   # If true, use modified linear layers in self-interaction block after the convolution in the interaction blocks.
    use_modal_output_block: True       # If true, use modified linear layer in the output block.
train:
    train_shuffle: True
    random_seed: 777
    is_train_stress : True
    epoch: 200

    loss: 'Huber'
    loss_param:
        delta: 0.01

    optimizer: 'adam'
    optim_param:
        lr: 0.01
    scheduler: 'linearlr'
    scheduler_param:
        start_factor: 1.0
        total_iters: 200
        end_factor: 0.0001

    force_loss_weight : 1.00
    stress_loss_weight: 0.01

    error_record:
        - ['Energy', 'MAE']
        - ['Force', 'MAE']
        - ['Stress', 'MAE']
        - ['Energy', 'Loss']
        - ['Force', 'Loss']
        - ['Stress', 'Loss']
        - ['TotalLoss', 'None']

    per_epoch: 10
    use_modality: True
    use_weight: True

data:
    batch_size: 16
    shift: 'elemwise_reference_energies'
    scale: 'force_rms'

    use_modal_wise_shift: True    # If true, use different atomic energy shift for each database
    use_modal_wise_scale: False   # If true, use different atomic energy scale for each database

    load_trainset_path:

      - data_modality: pbe  # Name of database
        file_list:
          - file: "path to pbe dataset"  # ASE readable or .pt file (graph.pt)
        data_weight:
            energy: 1.0
            force: 0.1   # This weight would be additionally multiplied to `force_loss_weight` for this database
            stress: 1.0  # This weight would be additionally multiplied to `stress_loss_weight` for this database

      - data_modality: scan
        file_list:
          - file: "path to scan dataset"
        data_weight:
            energy: 1.0
            force: 10.0
            stress: 1.0

    load_pbe_validset_path:   # any name starts with 'load' and ends with 'set_path'
      - data_modality: pbe   # modality must be given for mm valid set
        file_list:
          - file: "path to pbe validset"

    load_scan_validset_path:
      - data_modality: scan
        file_list:
          - file: "path to scan validset"