[ricardobeat]

It’s quite unlikely that training data will include duplicate repositories or even forks, that alone would surpass the published dataset sizes.