configuration_mxm.py

from transformers import PretrainedConfig
from typing import List


class MXMConfig(PretrainedConfig):
    model_type = "mxm"

    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__":
    mxm_config = MXMConfig(
        dim=128,
        n_layer=6,
        cutoff=5.0,
        num_spherical=7,
        num_radial=6,
        envelope_exponent=5,
        smiles=["C", "CC", "CCC"],
        processor_class="SmilesProcessor"
    )
    mxm_config.save_pretrained("custom-mxm")