modeling.py

from transformers import AutoConfig
from .modeling_custom import CustomModel

def from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs):
    # Load the configuration
    config = AutoConfig.from_pretrained(pretrained_model_name_or_path)
    
    # Fetch the number of emotion labels from the config
    num_emotion_labels = config.num_emotion_labels
    
    # Load the model with the correct configuration
    model = CustomModel.from_pretrained(
        pretrained_model_name_or_path,
        num_emotion_labels=num_emotion_labels,
        *model_args,
        **kwargs
    )
    return model