Datasets:

Modalities:
Text
Formats:
arrow
ArXiv:
Libraries:
Datasets
the-pile-splitted / README.md
ArmelR's picture
Update README.md
970296c
metadata
configs:
  - config_name: all
    data_files:
      - split: train
        path:
          - data/ArXiv/train/*.arrow
          - data/BookCorpus2/train/*.arrow
          - data/Books3/train/*.arrow
          - data/DM Mathematics/train/*.arrow
          - data/Enron Emails/train/*.arrow
          - data/EuroParl/train/*.arrow
          - data/FreeLaw/train/*.arrow
          - data/Github/train/*.arrow
          - data/Gutenberg (PG-19)/train/*.arrow
          - data/HackerNews/train/*.arrow
          - data/NIH ExPorter/train/*.arrow
          - data/OpenSubtitles/train/*.arrow
          - data/OpenWebText2/train/*.arrow
          - data/PhilPapers/train/*.arrow
          - data/Pile-CC/train/*.arrow
          - data/PubMed Abstracts/train/*.arrow
          - data/PubMed Central/train/*.arrow
          - data/StackExchange/train/*.arrow
          - data/UPSTO Backgrounds/train/*.arrow
          - data/Ubuntu IRC/train/*.arrow
          - data/Wikipedia (en)/train/*.arrow
          - data/YoutubeSubtitles/train/*.arrow
      - split: test
        path:
          - data/ArXiv/test/*.arrow
          - data/BookCorpus2/test/*.arrow
          - data/Books3/test/*.arrow
          - data/DM Mathematics/test/*.arrow
          - data/Enron Emails/test/*.arrow
          - data/EuroParl/test/*.arrow
          - data/FreeLaw/test/*.arrow
          - data/Github/test/*.arrow
          - data/Gutenberg (PG-19)/test/*.arrow
          - data/HackerNews/test/*.arrow
          - data/NIH ExPorter/test/*.arrow
          - data/OpenSubtitles/test/*.arrow
          - data/OpenWebText2/test/*.arrow
          - data/PhilPapers/test/*.arrow
          - data/Pile-CC/test/*.arrow
          - data/PubMed Abstracts/test/*.arrow
          - data/PubMed Central/test/*.arrow
          - data/StackExchange/test/*.arrow
          - data/UPSTO Backgrounds/test/*.arrow
          - data/Ubuntu IRC/test/*.arrow
          - data/Wikipedia (en)/test/*.arrow
          - data/YoutubeSubtitles/test/*.arrow
    default: true
  - config_name: ArXiv
    data_files:
      - split: train
        path: data/ArXiv/train/*.arrow
      - split: test
        path: data/ArXiv/test/*.arrow
  - config_name: BookCorpus2
    data_files:
      - split: train
        path: data/BookCorpus2/train/*.arrow
      - split: test
        path: data/BookCorpus2/test/*.arrow
  - config_name: Books3
    data_files:
      - split: train
        path: data/Books3/train/*.arrow
      - split: test
        path: data/Books3/test/*.arrow
  - config_name: DM Mathematics
    data_files:
      - split: train
        path: data/DM Mathematics/train/*.arrow
      - split: test
        path: data/DM Mathematics/test/*.arrow
  - config_name: Enron Emails
    data_files:
      - split: train
        path: data/Enron Emails/train/*.arrow
      - split: test
        path: data/Enron Emails/test/*.arrow
  - config_name: EuroParl
    data_files:
      - split: train
        path: data/EuroParl/train/*.arrow
      - split: test
        path: data/EuroParl/test/*.arrow
  - config_name: FreeLaw
    data_files:
      - split: train
        path: data/FreeLaw/train/*.arrow
      - split: test
        path: data/FreeLaw/test/*.arrow
  - config_name: Github
    data_files:
      - split: train
        path: data/Github/train/*.arrow
      - split: test
        path: data/Github/test/*.arrow
  - config_name: Gutenberg (PG-19)
    data_files:
      - split: train
        path: data/Gutenberg (PG-19)/train/*.arrow
      - split: test
        path: data/Gutenberg (PG-19)/test/*.arrow
  - config_name: HackerNews
    data_files:
      - split: train
        path: data/HackerNews/train/*.arrow
      - split: test
        path: data/HackerNews/test/*.arrow
  - config_name: NIH ExPorter
    data_files:
      - split: train
        path: data/NIH ExPorter/train/*.arrow
      - split: test
        path: data/NIH ExPorter/test/*.arrow
  - config_name: OpenSubtitles
    data_files:
      - split: train
        path: data/OpenSubtitles/train/*.arrow
      - split: test
        path: data/OpenSubtitles/test/*.arrow
  - config_name: OpenWebText2
    data_files:
      - split: train
        path: data/OpenWebText2/train/*.arrow
      - split: test
        path: data/OpenWebText2/test/*.arrow
  - config_name: PhilPapers
    data_files:
      - split: train
        path: data/PhilPapers/train/*.arrow
      - split: test
        path: data/PhilPapers/test/*.arrow
  - config_name: Pile-CC
    data_files:
      - split: train
        path: data/Pile-CC/train/*.arrow
      - split: test
        path: data/Pile-CC/test/*.arrow
  - config_name: PubMed Abstracts
    data_files:
      - split: train
        path: data/PubMed Abstracts/train/*.arrow
      - split: test
        path: data/PubMed Abstracts/test/*.arrow
  - config_name: PubMed Central
    data_files:
      - split: train
        path: data/PubMed Central/train/*.arrow
      - split: test
        path: data/PubMed Central/test/*.arrow
  - config_name: StackExchange
    data_files:
      - split: train
        path: data/StackExchange/train/*.arrow
      - split: test
        path: data/StackExchange/test/*.arrow
  - config_name: UPSTO Backgrounds
    data_files:
      - split: train
        path: data/UPSTO Backgrounds/train/*.arrow
      - split: test
        path: data/UPSTO Backgrounds/test/*.arrow
  - config_name: Ubuntu IRC
    data_files:
      - split: train
        path: data/Ubuntu IRC/train/*.arrow
      - split: test
        path: data/Ubuntu IRC/test/*.arrow
  - config_name: Wikipedia (en)
    data_files:
      - split: train
        path: data/Wikipedia (en)/train/*.arrow
      - split: test
        path: data/Wikipedia (en)/test/*.arrow
  - config_name: YoutubeSubtitles
    data_files:
      - split: train
        path: data/YoutubeSubtitles/train/*.arrow
      - split: test
        path: data/YoutubeSubtitles/test/*.arrow

Dataset description

The pile is an 800GB dataset of english text designed by EleutherAI to train large-scale language models. The original version of the dataset can be found here.

The dataset is divided into 22 smaller high-quality datasets. For more information each of them, please refer to the datasheet for the pile.

However, the current version of the dataset, available on the Hub, is not splitted accordingly. We had to solve this problem in order to improve the user experience when it comes to deal with the pile via the hub.

Here is an instance of the pile

{
  'meta': {'pile_set_name': 'Pile-CC'},
  'text': 'It is done, and submitted. You can play “Survival of the Tastiest” on Android, and on the web. Playing on...'
}

We used the meta column to properly divide the dataset in subsets. Each instance example belongs to the subset domain and domain = example['meta']['pile_set_name']. By doing this, we were able to create a new version of the pile that is properly divided, each instance having a new column domain.

We further splitted each subset in train/test (97%/3%) to build the current dataset which the following structure

data
  ArXiv
    train
    test
  BookCorpus2
    train
    test
  Books3
    train
    test

Usage

from datasets import load_dataset
dataset = load_dataset(
  "ArmelR/the-pile-splitted",
  subset_of_interest,
  num_proc=8
)

Using subset_of_interest = "default" will load the whole dataset.