No one is training their models on their closed sourced proprietary code. They own github, why would they need to do this.