The Pile An 800GB Dataset of Diverse Text for Language Modeling
EleutherAI

EleutherAI_ThePile_v1 (51 files)
README.txt 0.10kB
pile/SHA256SUMS.txt 2.78kB
pile/test.jsonl.zst 460.25MB
pile/train/00.jsonl.zst 15.24GB
pile/train/01.jsonl.zst 15.21GB
pile/train/02.jsonl.zst 15.21GB
pile/train/03.jsonl.zst 15.19GB
pile/train/04.jsonl.zst 15.19GB
pile/train/05.jsonl.zst 15.21GB
pile/train/06.jsonl.zst 15.26GB
pile/train/07.jsonl.zst 15.31GB
pile/train/08.jsonl.zst 15.23GB
pile/train/09.jsonl.zst 15.22GB
pile/train/10.jsonl.zst 15.23GB
pile/train/11.jsonl.zst 15.22GB
pile/train/12.jsonl.zst 15.26GB
pile/train/13.jsonl.zst 15.21GB
pile/train/14.jsonl.zst 15.22GB
pile/train/15.jsonl.zst 15.28GB
pile/train/16.jsonl.zst 15.27GB
pile/train/17.jsonl.zst 15.31GB
pile/train/18.jsonl.zst 15.31GB
pile/train/19.jsonl.zst 15.28GB
pile/train/20.jsonl.zst 15.21GB
pile/train/21.jsonl.zst 15.31GB
pile/train/22.jsonl.zst 15.30GB
pile/train/23.jsonl.zst 15.29GB
pile/train/24.jsonl.zst 15.19GB
pile/train/25.jsonl.zst 15.20GB
pile/train/26.jsonl.zst 15.20GB
pile/train/27.jsonl.zst 15.22GB
pile/train/28.jsonl.zst 15.22GB
pile/train/29.jsonl.zst 15.22GB
pile/val.jsonl.zst 470.91MB
pile_preliminary_components/2020-09-08-arxiv-extracts-nofallback-until-2007-068.tar.gz 17.48GB
pile_preliminary_components/EuroParliamentProceedings_1996_2011.jsonl.zst 1.48GB
pile_preliminary_components/FreeLaw_Opinions.jsonl.zst 17.01GB
pile_preliminary_components/Literotica.jsonl.zst 4.43GB
pile_preliminary_components/NIH_ExPORTER_awarded_grant_text.jsonl.zst 630.78MB
pile_preliminary_components/PMC_extracts.tar.gz 28.28GB
pile_preliminary_components/PUBMED_title_abstracts_2019_baseline.jsonl.zst 6.90GB
pile_preliminary_components/PhilArchive.jsonl.zst 797.71MB
pile_preliminary_components/books1.tar.gz 2.40GB
pile_preliminary_components/books3.tar.gz 39.52GB
pile_preliminary_components/github.tar 113.35GB
pile_preliminary_components/hn.tar.gz 706.52MB
pile_preliminary_components/openwebtext2.jsonl.zst.tar 29.34GB
pile_preliminary_components/pile_uspto.tar 11.79GB
pile_preliminary_components/stackexchange_dataset.tar 36.80GB
pile_preliminary_components/ubuntu_irc_until_2020_9_1.jsonl.zst 2.04GB
pile_preliminary_components/yt_subs.jsonl.zst 1.78GB
Type: Dataset
Tags:

Bibtex:
@article{,
title= {The Pile An 800GB Dataset of Diverse Text for Language Modeling},
journal= {},
author= {EleutherAI},
year= {},
url= {https://the-eye.eu/public/AI/},
abstract= {## What is the Pile?

The Pile is a 825 GiB diverse, open source language modelling data set that consists of 22 smaller, high-quality datasets combined together.

## Why is the Pile a good training set?
Recent work has shown that especially for large models, diversity in data sources improves general cross-domain knowledge of the model, as well as downstream generalization capability. In our evaluations, not only do models trained on the Pile show moderate improvements in traditional language modeling benchmarks, they also show significant improvements on Pile BPB.

## Why is the Pile a good benchmark?
To score well on Pile BPB (bits per byte), a model must be able to understand many disparate domains including books, github repositories, webpages, chat logs, and medical, physics, math, computer science, and philosophy papers. Pile BPB is a measure of world knowledge and reasoning ability in these domains, making it a robust benchmark of general, cross-domain text modeling ability for large language models.},
keywords= {},
terms= {},
license= {},
superseded= {}
}

Report