Host Deduped Pile raw jsonls

[haileyschoelkopf]

I don't think that the deduped Pile raw text data is hosted anywhere--I couldn't find it on the eye. Even if we don't host the deduped Pile .bin and .idx files somewhere, we definitely need to host the raw deduped Pile data to make these experiments replicable.

[aflah02]

@haileyschoelkopf Thanks for the great work on this repo! Is there any ETA for the Pile dataset hosting? Would be really helpful for some analysis

[haileyschoelkopf]

Hi, thanks for the interest!

I've uploaded the deduplicated Pile in Parquet format on HF datasets, which may be helpful:
https://huggingface.co/datasets/EleutherAI/the_pile_deduplicated

I'll try to upload the JSONLs today too, running into issues with filesize for our .bin and .idx files from NeoX.

Please let me know if you have any questions about replicating data ordering in training. We'll have something up on replicating the data seen by the model very soon!

[aflah02]

Thanks a ton, I'll reach out if I face some issues!

[haileyschoelkopf]

https://huggingface.co/datasets/EleutherAI/raw_deduplicated_pile BTW, all the deduped Pile jsonl files are here now :)

[aflah02]

Thanks!!