configuration_transmxm.py

from transformers import PretrainedConfig
from typing import List


class TransmxmConfig(PretrainedConfig):
    model_type = "transmxm"

    def __init__(
        self,
            dim: int = 128,
            n_layer: int = 6,
            cutoff: float = 5.0,
            num_spherical: int = 7,
            num_radial: int = 6,
            envelope_exponent: int = 5,

            smiles: List[str] = None,
            processor_class: str = "SmilesProcessor",
        **kwargs,
    ):
        self.dim = dim  # the dimension of input feature
        self.n_layer = n_layer  # the number of GCN layers
        self.cutoff = cutoff  # the cutoff distance for neighbor searching
        self.num_spherical = num_spherical  # the number of spherical harmonics
        self.num_radial = num_radial  # the number of radial basis
        self.envelope_exponent = envelope_exponent  # the envelope exponent

        self.smiles = smiles  # process smiles
        self.processor_class = processor_class


        super().__init__(**kwargs)


if __name__ == "__main__":
    transmxm_config = TransmxmConfig(
        dim=128,
        n_layer=6,
        cutoff=5.0,
        num_spherical=7,
        num_radial=6,
        envelope_exponent=5,
        smiles=["C", "CC", "CCC"],
        processor_class="SmilesProcessor"
    )
    transmxm_config.save_pretrained("custom-transmxm")