Sample from the Pile (EleutherAI/pythia_pile_idxmaps) used to compute the memorisation profiles.
The unique sequence identified (uid) is simply the order of the sequence in the Pile. The dataset is already tokenized.
Each file contains the same uid (for simplicity taken from pietrolesci/pile-deduped-subset), but since different seeds are used for data shuffling, the data are different.