Refer to TrOCR, which is an instance of [VisionEncoderDecoderModel].