Hugging Face's logo

Datasets:
ks48
/
url-atlas

Tasks:
Text Retrieval
Other
Modalities:
Text
Formats:
parquet
Languages:
multilingual
Size:
100B - 1T
Tags:
urls
web
common-crawl
wikipedia
reddit
hacker-news
Libraries:
Datasets
Dask
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
Dataset Preview
Duplicate
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
the dataset is currently locked, please try again later.
Error code:   LockedDatasetTimeoutError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

url
string
https://cec.aau.ac.ae/ar/units/admission-and-registration/contact-us
https://cbdreviewerp.web.app/wesep/318.html
https://cfy.ksu.edu.sa/ar/node/1044
https://crystalfoto.pl/%D8%A5%D8%B9%D8%A7%D8%AF%D8%A9/%D8%AA%D8%AF%D9%88%D9%8A%D8%B1/%D8%B3%D8%AD%D9%82/%D9%81%D9%8A/%D8%B9%D9%85%D9%84%D9%8A%D8%A9/%D8%A7%D9%84%D8%B7%D8%AD%D9%86/23230.html
https://cbdreviewphq.web.app/zyvih/cbd-699.html
https://decoration.7olm.org/t82-topic
https://defense-arab.com/vb/members/36498/ratings?reaction_category_id=1
https://chilchil.asia/Jul-29/4978.html
https://dinasoor.tech/tag/%D9%88%D8%B1%D8%AF-%D8%A8%D9%8A%D8%B1%D9%81%D9%83%D8%AA-%D8%A8%D8%B1%D9%86%D8%A7%D9%85%D8%AC-%D9%85%D8%B9%D8%A7%D9%84%D8%AC%D8%A9-%D9%86%D8%B5%D9%88%D8%B5/
https://chilchil.asia/Jul-17/24146.html
https://darek-bus.pl/27455/32RXhi1/%D9%83%D8%B3%D8%A7%D8%B1%D8%A9_%D8%A7%D9%84%D8%AD%D8%AC%D8%B1_%D8%A7%D9%84%D8%AC%D9%8A%D8%B1%D9%8A_%D8%A7%D9%84%D9%85%D8%AA%D9%86%D9%82%D9%84%D8%A9_%D8%A7%D9%84%D8%A3%D8%B1%D8%A8%D8%B9%D8%A7%D8%A1_%D9%84%D9%84%D8%A8%D9%8A%D8%B9_%D9%81%D9%8A_%D8%AC%D9%86%D9%88%D8%A8_%D8%A3%D9%81%D8%B1%...
https://diwanegypt.com/product/%D8%AA%D9%88%D9%82%D9%81-%D8%B9%D9%86-%D8%A7%D8%B1%D8%B6%D8%A7%D8%A1-%D8%A7%D9%84%D8%A7%D8%AE%D8%B1%D9%8A%D9%86/
https://ela5bar.com/2021/09/18/%D8%A7%D9%84%D8%A8%D8%B1%D9%84%D9%85%D8%A7%D9%86-%D8%A7%D9%84%D8%B9%D8%B1%D8%A8%D9%8A-%D9%8A%D8%A4%D9%83%D8%AF-%D8%B6%D8%B1%D9%88%D8%B1%D8%A9-%D8%A7%D9%84%D8%AA%D9%83%D8%A7%D8%AA%D9%81-%D8%A7%D9%84/
https://elibrary.mara.gov.om/mktbat-mezab/alkhzantt-alaamtt/ktab/?id=2906
https://elibrary.mara.gov.om/mktbat-mezab/alkhzantt-alaamtt/ktab/?id=3176
https://elibrary.mara.gov.om/mktbat-mezab/alkhzantt-alaamtt/ktab/?id=3473
https://elfagr.org/4072012
https://eljarida.com/news/5243
https://fawasildesign.com/zZWKRo
https://forum.rjeem.com/f13-35.html
https://elryad.com/ar/milo-portfolio/%D8%AA%D8%B5%D9%85%D9%8A%D9%85-%D9%85%D9%88%D9%82%D8%B9-%D8%A7%D8%AF%D9%88%D8%A7%D8%AA-%D8%B5%D8%AD%D9%8A%D8%A9/
https://g0ld0.com/product/xerjoff-join-the-club-don-u-edp-100-ml/
https://education-ksa.com/forumdisplay.php?s=79843aeaa5325499e650923c013453df&f=19
https://humum.net/?p=28316
https://halageorgia.com/%D9%82%D9%86%D8%A8%D9%84%D9%87-%D9%85%D9%88%D9%82%D9%88%D8%AA%D9%87-%D9%81%D9%8A-%D8%A8%D8%A7%D8%AA%D9%88%D9%85%D9%8A-%D8%A7%D9%84%D9%85%D9%81%D8%B1%D9%82%D8%B9%D8%A7%D8%AA-%D8%AA%D8%B9%D9%84%D9%86/
https://historiaeconomica.es/Barite-6152-JWBGJ2.html
https://hadarat.net/post/12167/%D8%AD%D9%85%D9%80-%D9%80%D8%A7%D8%B3-%D9%85%D8%AA%D9%85%D8%A7%D8%B3%D9%83%D8%A9-%D9%88%D9%81%D8%AA%D8%AD-%D9%85%D9%86%D9%82%D8%B3%D9%85%D8%A9
https://imazighen.univanet.com/private.php?s=d555dd3f5d41a71916508a8af3e446c5
https://iamblond.ru/%D9%81%D9%8A%D9%84%D9%85%20%D9%83%D8%B1%D8%AA%D9%88%D9%86%20%D8%B9%D9%84%D8%A7%D8%A1%20%D8%A7%D9%84%D8%AF%D9%8A%D9%86%20%D9%88%D8%B9%D9%88%D8%AF%D8%A9%20%D8%AC%D8%B9%D9%81%D8%B1%20%D9%85%D8%AF%D8%A8%D9%84%D8%AC%20%D8%B9%D8%B1%D8%A8%D9%8A
https://jadehiran.com/archives/tag/%D9%85%D8%AD%D9%85%D8%AF-%D8%B9%D9%84%D9%8A-%D9%86%D8%AC%D9%81%D9%8A
https://istefada.com/%D8%AA%D9%86%D8%B2%D9%8A%D9%84-%D9%88%D8%AA%D8%AD%D9%85%D9%8A%D9%84-%D9%83%D8%AA%D8%A7%D9%90%D8%A8-%D9%85%D9%86-%D9%8A%D8%A3%D9%83%D9%84-%D9%85%D9%84%D8%AD%D9%89-%D8%AA%D9%8A%D8%A7%D9%85%D8%A7%D8%AA/
https://jaidpub.org/
https://ida2at.org/news/2021/08/29/21022/------
https://jpfmjazan.com/%D9%83%D9%85_%D8%AA%D8%A7%D8%B1%D9%8A%D8%AE_%D8%A7%D9%84%D9%8A%D9%88%D9%85_%D8%A7%D9%84%D9%88%D8%B7%D9%86%D9%8A_%D8%A7%D9%84%D8%B3%D8%B9%D9%88%D8%AF%D9%8A_%D8%A8%D8%A7%D9%84%D9%87%D8%AC%D8%B1%D9%8A/
https://jelle-shop.com/%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD-%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD/c598890513
https://janoubia.com/2019/03/23/%D9%88%D8%B2%D8%A7%D8%B1%D8%A9-%D8%A7%D9%84%D9%85%D8%A7%D9%84-%D8%B1%D8%AF%D8%A7-%D8%B9%D9%84%D9%89-%D9%85%D8%A4%D8%B3%D8%B3%D8%A9-%D8%A7%D9%84%D9%83%D9%87%D8%B1%D8%A8%D8%A7%D8%A1-%D9%81%D8%AA%D8%AD/
ERROR: type should be string, got "https://kulalsalafiyeen.com/vb/tags.php?s=022fd3b13c5f639cdd03c648179e4e2f&tag=%C7%E1%DA%CB%ED%E3%ED%E4"
https://jan-violations.com/news/%D9%85%D9%82%D8%AA%D9%84-%D8%B9%D9%8A%D8%B3%D9%89-%D8%AD%D9%85%D9%8A%D8%AF%D9%8A-%D9%88%D8%A5%D8%B5%D8%A7%D8%A8%D8%A9-%D8%AE%D8%A7%D9%84%D8%AF-%D8%A7%D9%84%D8%A3%D8%B9%D8%B1%D8%A7%D8%AC-%D9%88/
https://kuwaityiat.net/blog/category/%D8%BA%D8%B3%D8%A7%D9%84%D8%A7%D8%AA/
https://kulalsalafiyeen.com/vb/misc.php?s=2045804357a6cdef3f5279149db3c246&do=bbcode
https://kuwaityiat.net/blog/tag/%D8%B5%D9%8A%D8%A7%D9%86%D8%A9-%D8%AA%D9%83%D9%8A%D9%8A%D9%81-%D9%85%D8%B1%D9%83%D8%B2%D9%8A-%D8%A7%D9%84%D8%AF%D8%B9%D9%8A%D8%A9/
https://locateinkuwait.com/%D8%A3%D8%B9%D9%84%D8%A7%D9%86%D8%A7%D8%AA/%D9%85%D9%87%D9%86-%D9%88%D9%85%D9%82%D8%A7%D9%88%D9%84%D8%A7%D8%AA/194-%D9%84%D9%84%D8%A8%D9%8A%D8%B9-%D8%B4%D9%82%D8%A9-%D8%AA%D9%85%D9%84%D9%8A%D9%83-%D8%AD%D8%AF%D8%A7%D8%A6%D9%82-%D8%A7%D9%84%D9%82%D8%A8%D8%A9-%D8%B4-%D8%A8%D9%88%D8%B1-%D8%B3%D8...
https://m.iium.edu.my/ar/news/talk-organized-by-iium-siber-sejahtera-and-bank-negara-malaysia-on-financial-crime-awareness
https://makhtota.ksu.edu.sa/makhtota/2741/8
https://makhtota.ksu.edu.sa/makhtota/2545/6
https://maintenance-company-emirates.com/tag/%D9%85%D8%B5%D8%A7%D9%86%D8%B9-%D8%AD%D8%AC%D8%B1-%D8%B7%D8%A8%D9%8A%D8%B9%D9%8A/
https://dev1.kataeb.org/%D8%B9%D8%A8%D8%AF%D8%A7%D9%84%D9%84%D9%87-%D8%A7%D9%86%D9%81%D8%AA%D8%A7%D8%AD-%D8%A7%D9%84%D8%AF%D9%88%D9%84-%D8%A7%D9%84%D8%B9%D8%B1%D8%A8%D9%8A%D8%A9-%D8%B9%D9%84%D9%89-%D9%84%D8%A8%D9%86%D8%A7%D9%86-%D9%84%D9%8A%D8%B3-%D9%82%D8%B1%D9%8A%D8%A8%D8%A7/2021/10/14/%D9%85%D8%AD%D9%84%D9%8A%D8%A7%...
https://lingolets.com/arabic-to-english/%D8%A3%D8%AC%D8%B1%D9%89
https://maktabeti.com/tag/%D8%B1%D8%A7%D8%AA%D8%A8/
https://mart.ps/325-leather-watches
https://midad.com/recitation/201375/%D8%B3%D9%88%D8%B1%D8%A9-%D8%A7%D9%84%D8%B9%D9%86%D9%83%D8%A8%D9%88%D8%AA
https://mediacenter.ps/archives/tag/%D8%A7%D9%84%D8%B1%D8%A7%D8%AD%D9%84-%D8%A8%D9%88%D8%B1%D9%82%D9%8A%D8%A8%D8%A9
https://media.sn4hr.org/blog/2019/04/30/59569/
https://madany.ahlamontada.com/t6893-topic
https://mawdea.com/28606/%D8%B1%D8%B3%D8%A7%D8%A6%D9%84-%D8%B1%D9%88%D9%85%D8%A7%D9%86%D8%B3%D9%8A%D8%A9-%D9%85%D8%B3%D8%A7%D8%A6%D9%8A%D8%A9-%D9%84%D8%AD%D8%A8%D9%8A%D8%A8%D8%AA%D9%8A/
https://misralbalad.com/%D8%B9%D8%B7%D9%84-%D9%85%D9%81%D8%A7%D8%AC%D8%A6-%D8%A8%D9%80-%D9%81%D9%8A%D8%B3-%D8%A8%D9%88%D9%83-%D9%88%D9%88%D8%A7%D8%AA%D8%B3%D8%A7%D8%A8-%D9%88%D8%A7%D9%86%D8%B3%D8%AA%D8%AC%D8%B1%D8%A7%D9%85/
https://misralbalad.com/338585-2/
https://mirosaintgermainenlaye.fr/%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD-%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD/23935/%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD.html
https://mnoaat.com/%D9%88%D8%B5%D9%81%D8%A7%D8%AA-%D8%B7%D8%A8%D8%AE/1868/%D8%B7%D8%B1%D9%8A%D9%82%D8%A9-%D8%B9%D9%85%D9%84-%D8%A7%D9%84%D9%83%D8%B4%D8%B1%D9%89/
https://nala4u.com/2021/02/20/%D9%87%D8%AF%D9%86%D8%A9-%D9%85%D8%B9-%D8%A7%D9%84%D8%B0%D8%A7%D8%AA/
https://new.yallashootkora.com/2021/04/21-4-2021_4.html
https://news.un.org/ar/tags/technical-migrate/date/2015-03
https://oudah.net/%D9%81%D9%8A%D8%AA%D9%86%D8%A7%D9%85%D9%8A-%D8%AC%D8%A8%D9%84%D9%8A-%D8%B3%D9%88%D8%A8%D8%B1/p358777781
https://poetsgate.com/poem.php?pm=211553&name=%D8%B9%D9%84%D8%A7%D8%A1%20%D8%A7%D9%84%D8%A3%D8%AF%D9%8A%D8%A8&Title=%D9%85%D9%86%20%D9%82%D8%A7%D9%84%20%D8%A7%D9%86%D9%8A%20%D8%B0%D8%A7%D8%AA%20%D9%8A%D9%88%D9%85%20%D8%A7%D8%B9%D8%B4%D9%82
https://obsdarwin.nl/Jun-03/33883.html
https://projectssolutions.onepeterson.com/ar/industries/agriculture
https://optionsjehi.web.app/badini34274qyz/dyp.html
https://optionsjehi.web.app/gruska18329rapu/xuva.html
https://ramadan2021.alarab.com/v107715-_%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD_29_%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD_%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD_HD_292020
https://recas.ru/ar/universities-of-russia/sports-universities/582-far-eastern-state-academy-of-physical-culture
https://rewayat.club/novel/elite-cultivater
https://ramadan2021.alarab.com/v112487-_%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD_129_%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD_%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD_%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF%BF%BD%EF...
https://osp-kwiatowe.pl/02_Dec_2006_2625.html
https://salleeshop.com/product/dress-pimkie/
https://saudishift.com/media/736219985465783376
https://sa.made-in-china.com/co_rainbowtraco/product_PDA-LCD-Display-for-Mio-136-138-139-268-269-336-339-558-NL2432HC22-23B_hehooeirg.html
https://sollywood.com.sa/2018/05/06/a-215/
https://st-takla.org/General-Knowledge-Articles/07-Calendars/1200-1299/year-1277.html
https://sanews.pythonanywhere.com/post/954/
https://st-takla.org/Saints/Coptic-Orthodox-Saints-Biography/Coptic-Saints-Story_774.html
https://ssrcaw.org/ar/show.art.asp?t=2&aid=731994
https://st-takla.org/books/fr-tadros-malaty/patristic-social-line/wars.html
https://ralqalam.com/article/%D9%85%D9%8E%D8%AB%D9%8E%D9%84%D9%8F-%D8%A7%D9%84%D9%83%D9%84%D9%85%D8%A9-%D8%A7%D9%84%D8%B7%D9%8A%D8%A8%D8%A9-%D9%88%D8%A7%D9%84%D9%83%D9%84%D9%85%D8%A9-%D8%A7%D9%84%D8%AE%D8%A8%D9%8A%D8%AB%D8%A9-%D9%81%D9%8A-%D8%B3%D9%88%D8%B1%D8%A9-%D8%A5%D8%A8%D8%B1%D8%A7%D9%87%D9%8A%D9%85
https://stars-apps.com/
https://taqadoumi.net/2020/07/11/%D8%B9%D8%A7%D8%AC%D9%84-%D8%B1%D8%A6%D9%8A%D8%B3-%D9%84%D8%AC%D9%86%D8%A9-%D8%A7%D9%84%D8%B9%D8%AF%D9%84-%D9%8A%D8%B1%D8%AF-%D8%B9%D9%84%D9%89-%D8%AA%D8%B5%D8%B1%D9%8A%D8%AD%D8%A7%D8%AA-%D8%A8/
https://sudanewsnow.com/159075/
https://st-takla.org/books/pope-sheounda-iii/resurrection/message.html
https://studiosday.web.app/ide-10-ja.html
https://torath.gov.ae/module/branches/all/3
https://tv.sayidaty.net/node/11996/%D8%A3%D8%B2%D9%8A%D8%A7%D8%A1/%D8%A7%D9%84%D8%A8%D9%84%D9%88%D8%BA%D8%B1-%D9%83%D8%B1%D9%8A%D8%B3%D8%AA%D9%8A%D9%86%D8%A7-%D9%83%D8%B1%D9%85-%D8%B1%D9%85%D9%8A%D8%A7-%D8%AA%D9%81%D8%AA%D8%AD-%D8%A3%D8%A8%D9%88%D8%A7%D8%A8-%D8%AE%D8%B2%D8%A7%D9%86%D8%AA%D9%87%D8%A7-%D9%88%D8%AA%D9%83%...
https://thezunguness.co.za/%D8%B5%D9%88%D8%B1/%D9%85%D8%AD%D8%A7%D9%85%D9%84/%D9%83%D8%B3%D8%A7%D8%B1%D8%A9/%D8%A7%D9%84%D9%81%D9%83/39749.html
https://sudanewsnow.com/157293/
https://theaqd.com/%D8%B5%D9%8A%D8%BA%D8%A9-%D8%A7%D9%84%D8%B9%D9%82%D8%AF-%D9%81%D9%8A-%D8%A7%D9%84%D8%B9%D9%84%D8%A7%D9%82%D8%A7%D8%AA-%D8%A7%D9%84%D8%AA%D8%B9%D8%A7%D9%82%D8%AF%D9%8A%D8%A9/
https://topbinibrh.web.app/bumpaus11151xek/rywa.html
https://wafaamagazine.org/archives/11300
https://www.adsgal.com/author/evodak-konaklama/?type=ads
https://www.adsgal.com/search-results/?cat_id=102
https://www.adtv.ae/video/27679972/Miss-Understand
https://video.marebpress.net/articles.php?id=42319
https://www.218tv.net/%D8%AA%D8%A3%D9%8A%D9%8A%D8%AF-%D9%8A%D9%88%D9%86%D8%A7%D9%86%D9%8A-%D9%84%D9%85%D8%B9%D8%A7%D8%B1%D8%B6%D8%A9-%D8%A7%D9%84%D9%86%D9%88%D8%A7%D8%A8-%D8%B9%D9%84%D9%89-%D9%85%D8%B0%D9%83%D8%B1/
End of preview.

URL Atlas

257,548,097,528 URLs from 105 web corpora, each kept as its own separately-loadable config, plus the raw source dumps two of them were extracted from. 4.35 TiB across 52,244 files.

This is the input side of a URL-compression corpus: every source reduced to its URL column and nothing else. It is deliberately not deduplicated or merged — sources are kept intact and overlapping so you can measure what each one contributes, pick the subset you want, and dedup on your own terms.

Quick start

Nothing here is small, so always name a config — there is no default, which is the point:

from datasets import load_dataset

ds = load_dataset("ks46/url-atlas", "wiki-urls", split="train", streaming=True)
next(iter(ds))          # {'url': 'https://...'}

Grab one source without the datasets library:

hf download ks46/url-atlas --repo-type dataset \
    --include 'data/hackernews-urls/*' --local-dir ./url-atlas

Query it in place with DuckDB:

SELECT url FROM read_parquet(
  'hf://datasets/ks46/url-atlas/data/finemath_urls_4plus/*.parquet') LIMIT 10;

What is in here

URLs sources
Original extractions (not published anywhere else) 5,647,739,601 6
Common Crawl columnar-index sweep 123,873,807,225 47
Public corpora projected to their URL column 128,026,550,702 52
total 257,548,097,528 105

Layout 

data/<source>/*.parquet     one nullable `url` string column, zstd
raw/wiki/*.sql.gz           890 Wikimedia `externallinks` SQL dumps
raw/hackernews/*.parquet    49,268,259 Hacker News items, full schema
source-notes/<source>.md    provenance and preprocessing, one file per source

raw/ is the unprocessed source for two of the extractions, kept so the extraction is reproducible and auditable rather than something you have to take on faith. raw/wiki is the input to data/wiki-urls; raw/hackernews is the input to data/hackernews-urls and is a config in its own right (hackernews-raw) since it carries titles, scores, timestamps and comment text.

The original extractions

These exist nowhere else and are the reason this is a dataset rather than a mirror:

source URLs files on disk derived from
reddit-submissions 3,210,035,355 5783 37.6 GiB open-index/arctic
reddit-comments 1,341,471,971 26031 15.6 GiB open-index/arctic
wiki-urls 982,404,166 1039 10.1 GiB —
reddit-wenknow 80,696,485 797 1.3 GiB wenknow/reddit_dataset_888
archive-text-urls 24,049,508 119 0.1 GiB Navanjana/ARCHIVE-TEXT-URLS
hackernews-urls 9,082,116 2 0.1 GiB —
  • wiki-urls — every outbound link in 1,087 Wikimedia wikis, not just English, parsed from the externallinks SQL dumps. Since MediaWiki 1.41 that table stores a reversed-domain index plus path rather than the URL an editor typed, so these are canonical reconstructions: host lowercased, userinfo gone.
  • reddit-comments / reddit-submissions — URLs scraped out of comment markdown and out of submissions (the posted link plus links in the self-text). 1.05 TiB of comment bodies reduce to the links inside them.
  • hackernews-urls — submitted links plus URLs found in comment HTML.
  • archive-text-urls — URLs scraped from plain-text archives.

Extraction is regex-based and three details do most of the damage if you get them wrong, so they are worth stating: balanced parentheses are kept (drop them and /wiki/Mercury_(element) silently becomes the real-but-different /wiki/Mercury; 8.4% of Wikipedia URLs contain parens); HTML entities are decoded before trailing punctuation is trimmed (trim first and &amp; leaves &amp welded on); and a closing bracket is removed only when the URL contains no opener.

The Common Crawl sweep

47 crawls of Common Crawl's columnar index, read directly rather than via a derived corpus. Novelty is concentrated in recent crawls — 2014 contributes 19–25% new URLs against the rest of this dataset while 2025 contributes 62–65%, because the derived corpora below were built from historical snapshots and none of them covers 2025.

source URLs files on disk derived from
cc-2014-42 3,722,131,678 300 14.3 GiB Common Crawl columnar index CC-MAIN-2014-42
cc-2014-23 3,608,720,876 300 17.6 GiB Common Crawl columnar index CC-MAIN-2014-23
cc-2022-21 3,449,824,215 300 35.4 GiB Common Crawl columnar index CC-MAIN-2022-21
cc-2023-40 3,445,015,037 300 36.5 GiB Common Crawl columnar index CC-MAIN-2023-40
cc-2022-49 3,381,612,888 300 35.0 GiB Common Crawl columnar index CC-MAIN-2022-49
cc-2023-50 3,354,042,124 300 35.6 GiB Common Crawl columnar index CC-MAIN-2023-50
cc-2018-30 3,263,958,200 300 31.4 GiB Common Crawl columnar index CC-MAIN-2018-30
cc-2018-13 3,232,738,018 300 31.0 GiB Common Crawl columnar index CC-MAIN-2018-13
cc-2023-06 3,189,578,207 300 33.4 GiB Common Crawl columnar index CC-MAIN-2023-06
cc-2022-40 3,176,763,048 300 33.1 GiB Common Crawl columnar index CC-MAIN-2022-40
cc-2017-04 3,146,467,312 300 25.5 GiB Common Crawl columnar index CC-MAIN-2017-04
cc-2021-17 3,134,424,806 300 31.8 GiB Common Crawl columnar index CC-MAIN-2021-17
cc-2023-14 3,118,633,413 300 32.8 GiB Common Crawl columnar index CC-MAIN-2023-14
cc-2022-27 3,109,609,672 300 32.3 GiB Common Crawl columnar index CC-MAIN-2022-27
cc-2024-10 3,106,525,566 300 33.7 GiB Common Crawl columnar index CC-MAIN-2024-10
cc-2018-26 3,069,414,247 300 29.1 GiB Common Crawl columnar index CC-MAIN-2018-26
cc-2018-43 3,041,784,523 300 29.3 GiB Common Crawl columnar index CC-MAIN-2018-43
cc-2025-05 3,031,278,337 300 33.2 GiB Common Crawl columnar index CC-MAIN-2025-05
cc-2017-22 2,967,174,165 300 25.9 GiB Common Crawl columnar index CC-MAIN-2017-22
cc-2017-17 2,942,879,945 300 25.2 GiB Common Crawl columnar index CC-MAIN-2017-17
cc-2017-51 2,914,004,423 300 27.5 GiB Common Crawl columnar index CC-MAIN-2017-51
cc-2024-26 2,798,047,026 300 31.0 GiB Common Crawl columnar index CC-MAIN-2024-26
cc-2024-18 2,786,800,057 300 30.7 GiB Common Crawl columnar index CC-MAIN-2024-18
cc-2021-10 2,736,749,883 300 28.1 GiB Common Crawl columnar index CC-MAIN-2021-10
cc-2020-45 2,731,063,243 300 27.5 GiB Common Crawl columnar index CC-MAIN-2020-45
cc-2018-34 2,686,323,358 300 26.1 GiB Common Crawl columnar index CC-MAIN-2018-34
cc-2025-08 2,679,706,056 300 29.8 GiB Common Crawl columnar index CC-MAIN-2025-08
cc-2014-15 2,641,371,316 300 14.3 GiB Common Crawl columnar index CC-MAIN-2014-15
cc-2024-51 2,635,362,356 300 29.4 GiB Common Crawl columnar index CC-MAIN-2024-51
cc-2021-21 2,632,142,465 300 27.2 GiB Common Crawl columnar index CC-MAIN-2021-21
cc-2025-43 2,616,796,857 300 29.8 GiB Common Crawl columnar index CC-MAIN-2025-43
cc-2022-33 2,588,690,250 300 27.5 GiB Common Crawl columnar index CC-MAIN-2022-33
cc-2019-39 2,561,062,912 300 25.9 GiB Common Crawl columnar index CC-MAIN-2019-39
cc-2021-25 2,457,631,756 300 26.1 GiB Common Crawl columnar index CC-MAIN-2021-25
cc-2020-34 2,449,834,239 300 24.5 GiB Common Crawl columnar index CC-MAIN-2020-34
cc-2025-38 2,385,964,209 300 27.4 GiB Common Crawl columnar index CC-MAIN-2025-38
cc-2025-47 2,294,472,912 300 26.3 GiB Common Crawl columnar index CC-MAIN-2025-47
commoncrawl 2,149,001,456 300 25.4 GiB Common Crawl columnar index CC-MAIN-2026-30
cc-2015-18 2,115,818,059 300 14.5 GiB Common Crawl columnar index CC-MAIN-2015-18
ccrawl-urls 2,098,491,742 300 24.8 GiB open-index/ccrawl-urls
cc-2026-12 1,974,845,234 300 23.0 GiB Common Crawl columnar index CC-MAIN-2026-12
cc-2015-35 1,848,022,475 300 12.8 GiB Common Crawl columnar index CC-MAIN-2015-35
cc-2015-48 1,824,170,527 300 12.6 GiB Common Crawl columnar index CC-MAIN-2015-48
cc-2016-22 1,466,220,798 300 10.9 GiB Common Crawl columnar index CC-MAIN-2016-22
cc-2016-26 1,236,815,660 300 8.8 GiB Common Crawl columnar index CC-MAIN-2016-26
cc-filtered-urls 67,479,353 606 0.4 GiB xinyangli/cc-filtered-urls
common-crawl-sample_urls 4,342,326 1 0.1 GiB nhagar/common-crawl-sample_urls

Public corpora, projected to their URL column

Mostly built on nhagar's *_urls extracts, which publish url + domain; we keep url and drop domain. These overlap each other heavily by construction — every Common-Crawl-derived corpus shares a backbone — which is why they are kept separate instead of merged.

source URLs files on disk derived from
fineweb 24,505,935,751 101 617.8 GiB nhagar/fineweb_urls
hplt2.0-cleaned 10,558,904,807 59 228.5 GiB nhagar/hplt2.0_cleaned_urls
colossal-oscar-1.0 10,332,304,603 52 250.6 GiB nhagar/colossal-oscar-1.0_urls
c4-multilingual 8,522,562,521 1104 244.1 GiB nhagar/c4_urls_multilingual
culturax 7,184,124,703 36 182.3 GiB nhagar/culturax_urls
madlad-400-noisy 6,801,149,222 25 193.8 GiB nhagar/madlad-400_urls_noisy
redpajama-data-v2 6,079,335,069 35 149.7 GiB nhagar/redpajama-data-v2_urls
txt360_urls 5,651,306,105 97 141.6 GiB nhagar/txt360_urls
dolma-v1.5 4,921,549,644 32 127.0 GiB nhagar/dolma_urls_v1.5
hplt-v1.2 4,651,480,080 7 60.6 GiB nhagar/hplt-v1.2_urls
fineweb-2 4,567,627,672 18 109.8 GiB nhagar/fineweb-2_urls
zyda-2 4,506,333,003 866 111.6 GiB nhagar/zyda-2_urls
dolma-v1.6 3,731,835,146 30 92.4 GiB nhagar/dolma_urls_v1.6
madlad-400-clean 3,507,642,707 14 96.2 GiB nhagar/madlad-400_urls_clean
dclm-baseline-1.0 2,949,254,346 15 72.3 GiB nhagar/dclm-baseline-1.0-parquet_urls
olmoe-mix-0924_urls 2,948,096,911 15 73.5 GiB nhagar/olmoe-mix-0924_urls
dolma_urls_v1.7 2,673,664,597 18 66.7 GiB nhagar/dolma_urls_v1.7
redpajama-data-1t_urls 1,469,452,453 11 37.8 GiB nhagar/redpajama-data-1t_urls
oscar-2301_urls 1,401,594,726 8 35.2 GiB nhagar/oscar-2301_urls
lucie-training-dataset_urls 1,281,707,164 12 32.2 GiB nhagar/lucie-training-dataset_urls
oscar-2109_urls 1,239,027,220 1 31.0 GiB nhagar/oscar-2109_urls
culturay_urls 1,220,338,131 7 18.2 GiB nhagar/culturay_urls
zyda_urls 1,174,929,887 105 29.9 GiB nhagar/zyda_urls
oscar-2201_urls 992,957,761 6 25.8 GiB nhagar/oscar-2201_urls
falcon-refinedweb_urls 968,000,015 95 25.0 GiB nhagar/falcon-refinedweb_urls
onlysports_dataset_urls 863,989,915 4 22.9 GiB nhagar/onlysports_dataset_urls
glotcc-v1_urls 684,697,321 5 16.5 GiB nhagar/glotcc-v1_urls
c4_urls_en.noblocklist 393,391,519 197 10.1 GiB nhagar/c4_urls_en.noblocklist
fineweb2-hq_urls 380,138,261 13 9.1 GiB nhagar/fineweb2-hq_urls
c4_en_urls 365,233,500 61 9.4 GiB nhagar/c4_en_urls
moscar_urls 303,229,385 2 7.1 GiB nhagar/moscar_urls
c4_urls_en.noclean 189,396,098 95 5.2 GiB nhagar/c4_urls_en.noclean
varta-urls 177,768,440 66 4.6 GiB rahular/varta-urls
mc4-es-sampled_urls 150,000,000 1 4.2 GiB nhagar/mc4-es-sampled_urls
obelics_urls 141,047,697 1 4.1 GiB nhagar/obelics_urls
fulg_urls 114,679,939 1 2.5 GiB nhagar/fulg_urls
clean_mc4_it_urls 101,631,883 1 2.8 GiB nhagar/clean_mc4_it_urls
culturax-mini-nonshuffled_urls 71,841,248 1 1.8 GiB nhagar/culturax-mini-nonshuffled_urls
mc4_nl_cleaned_urls 64,545,895 1 1.6 GiB nhagar/mc4_nl_cleaned_urls
catalog_urls 34,314,510 1 0.6 GiB nhagar/catalog_urls
101_billion_arabic_words_dataset_urls 33,059,988 1 0.9 GiB nhagar/101_billion_arabic_words_dataset_urls
infimm-webmath-40b_urls 23,739,397 1 0.5 GiB nhagar/infimm-webmath-40b_urls
finemath_urls_3plus 21,405,610 1 0.5 GiB nhagar/finemath_urls_3plus
mc4_fi_cleaned_urls 18,160,122 1 0.4 GiB nhagar/mc4_fi_cleaned_urls
c4_urls_realnewslike 13,799,838 7 0.4 GiB nhagar/c4_urls_realnewslike
crawlpt_dedup_urls 10,888,966 1 0.3 GiB nhagar/crawlpt_dedup_urls
finemath_urls_4plus 6,699,493 1 0.1 GiB nhagar/finemath_urls_4plus
open-web-math_urls 6,315,233 1 0.1 GiB nhagar/open-web-math_urls
georgian-corpus_urls 5,333,536 1 0.1 GiB nhagar/georgian-corpus_urls
legal-mc4_urls 4,934,201 1 0.1 GiB nhagar/legal-mc4_urls
c4-chinese-zhtw_urls 2,967,556 1 0.1 GiB nhagar/c4-chinese-zhtw_urls
mixturevitae-fineweb-permissive-multilingual-2m_urls 2,226,907 1 0.0 GiB nhagar/mixturevitae-fineweb-permissive-multilingual-2m_urls

Format, and two honest caveats

Every file under data/ is a single nullable url string column compressed with zstd. Row counts above are from a footer scan of all 52,244 files taken the day this was published, not from the upstream READMEs.

1. Two Parquet flavours, and no guaranteed row order. Write settings changed partway through staging and existing files were deliberately not rewritten. About half the sources are Parquet 2.6 with DELTA_LENGTH_BYTE_ARRAY; the rest are Parquet 1.0 with PLAIN, which costs roughly twice the bytes for the same URLs. Both read identically — read_parquet('data/x/*.parquet') does not care.

Row order is not guaranteed — do not rely on it. Files were originally written ORDER BY surt(url), url, but a later in-place repair pass rewrote values (decoding &amp; to &) after they had been sorted, and changing the string changes its SURT key. Checked against the reference SURT implementation, most sources are now partially out of order; rows containing & are about three times over-represented among the out-of-order ones, which is the fingerprint of that repair. Sources that never needed the repair are largely still ordered. If you need an ordering, impose it yourself — one ORDER BY surt(url), url does it.

This also costs bytes: measured on 6.7M URLs, unsorted is 146–147 MB against 74 MB sorted, so these files carry roughly twice the bytes a sorted copy would. They are published as-is rather than rewritten, so that what you download is byte-identical to what the build actually read.

2. Nothing is deduplicated, within or across sources. The counts above are pre-dedup and the sources genuinely overlap. On a 1/1024 consistent-hash sample of the whole corpus the unique fraction came out at 27.4%. Treat the total as "URLs staged", not "distinct URLs". Three pairs that were byte-identical republications (cultura/culturax, falcon/falcon-refinedweb, c4_urls_en/c4_en_urls) were already removed before publication, reclaiming 218.6 GiB with no information lost.

Where files are sorted, the key is ORDER BY surt(url), url using a byte-exact port of internetarchive/surt, so shards can be compared against each other and joined against Common Crawl's url_surtkey. The secondary sort on url matters because surt() is not injective — http/https and trailing-slash variants collapse to one key.

Licensing and attribution

This collection is a derivative work and carries no single license. Each config inherits the terms of the corpus it was extracted from — see the derived from column above and source-notes/<source>.md for the exact upstream, then follow that upstream's license. Notable cases: the Common Crawl index is under Common Crawl's terms of use; Wikimedia dumps are CC BY-SA / GFDL; most nhagar *_urls extracts inherit ODC-By from C4, FineWeb and friends.

URLs are addresses, not content: nothing here reproduces the text of any page. Some URLs will nonetheless point at content that is offensive, illegal in your jurisdiction, or personally identifying, and some will be dead. There is no safety filtering of any kind.

Provenance

Each source is fetched whole (the Hugging Face Xet client sustains ~359 MB/s on large files, where ranged column reads run below 50 MB/s and get rate-limited), projected to its url column with DuckDB, and the source file deleted — about 7.5 TiB pulled over the wire becomes the 4.35 TiB kept here.
Downloads last month
455
Number of rows:
222,565,313,863
Total file size:
7.35 TB