Fine-tuning the model is not supported yet
According to the paper, textual input should be also supported.