A refined version of PubMed Central dataset in The Pile by Data-Juicer. Removing some "bad" samples from the original dataset to make it higher-quality.
This dataset is usually used to pretrain a Large Language Model.
Notice: Here is a small subset for previewing. The whole dataset is available here (About 83G).
Number of samples: 2,694,860 (Keep ~86.96% from the original dataset)… See the full description on the dataset page:
https://huggingface.co/datasets/datajuicer/the-pile-pubmed-central-refined-by-data-juicer.
