About Pythia Deduped Models' Dataset

[Harperrrrr]

Hi,
you guys are really doing a wonderful job and thanks a lot for that! I have noticed that the pythia and pythia-deduped models are actually trained on different datasets (which are the Pile and the pile-deduplicated), but I could not find the steps to differentiate between the two in the code. Did I miss something or they just share the same dataset in the code? And I was wondering if your dataset on HuggingFace is entirely from the Pile, or if it includes the pile-deduplicated dataset?