Custom pre-trained model does not have config, can't use pipelines

I recently trained a Roberta tokenizer and MLM model from scratch and saved the keras model using MLFlow. When I load the model back I have no .config associated with the model, and therefore continue to get this error. Is there a workaround here, or a way to add the config variable after loading the model? There is a get_config() method available that seems to provide the config dictionary.