This worker will be used for smol datasets (to avoid doing the corpus export which can take a while).