They are not exporting the models, they are exporting very speech like output.