Several smaller versions of the models have been trained on the same dataset.