No, it should be just the amount of files remaining

Hi @<1547028031053238272:profile|MassiveGoldfish6> , yes, every new version contains all included files

so if my parent dataset is 1Tb and I add a single file to create a child dataset. There will now be 2Tb of data on the server. The parent dataset is duplicated on the server?

is this still true if the child dataset is smaller than the parent? If the parent dataset is 1Tb and I delete half the files, I will still be pushing 2Tb of data to the server?

Yes. Differential Datasets are part of the ClearML Scale and Enterprise solution 😞