GUIDE de sociétés

LAION ak ay done yu ubbeeku

LAION kuréel la guy def te Yàlla tax, génne ay done yu bari yu ubbeeku ci nataal, li ci gëna siiw mooy LAION-5B, moo jàppale tàggat model yu ubbeeku yu melni Stable Diffusion.

2 simili jàngDañu mujjee yeesal

Résumé

It matters because it made web-scale multimodal data freely available to researchers outside large corporations.

Plongeur bu xóot

LAION (Reseau bu ubbeeku ci xarañteg xelu masin) kuréel la guñ taxawal ci atum 2021 ngir demokraatise gëstu masin ci génne ay done yu ubbeeku. Li gëna siiw ci limu génne, LAION-5B, amna lu tollu ci 5.85 milyaari nataal-tekst yuñ segg ci done web Common Crawl di jëfandikoo xeetu CLIP bu OpenAI ngir tëye ñaar yi ci barab bi mbind mi ak nataal bi méngoo. Li gëna am solo mooy LAION du jël nataal yi ci seen bopp; dafay séddale ay URL ak ay metadata, suko defee jëfandikukat yi yebbi nataal yi ci balluwaayi web yi. Done yooyu dañu amal njariñ ci tàggat Stable Diffusion ak yeneen xeeti mbind yu ubbeeku ci nataal. LAION dafa jànkoonte ak saytu bu tar: ci 2023 gëstukat yi gis nañu lëkkalekaay ci nataali yàqu-yàqu budul yoon ci done yi, loolu waral LAION mu wàcce ko, setal ko, ba noppi gennewaat benn xeetu gëna wóor, di wane risku yi ci scraping web-scale bu filtrewul.

Gis-gis xarala

LAION-5B dañu ko tabax ci skane Common Crawl ngir am etiketu nataal HTML ak alt-text, ginaaw ga ñu jëfandikoo CLIP ngir xayma nuru gi am ci digganté nataal bu nekk ak liñu ci bind. Peer yi nekk ci suufu cosine-similarity threshold dañu leen sànni, kon peer nataal-tekst yu méngoo bu baax kese ñoo desoon. Done yi dañu leen xaaj làkk yi, ba noppi dañu amaale ay CLIP yuñ njëkka xayma, loolu mooy tax ñu mëna seetlu lu gaaw ci nuru. Ndax URL yi kese lañuy denc, lëkkalekaay bi dafay yàq ndànk-ndànk mën nañu ko defaraat lu yàgg.

njeextalu pexe

Pexem jaaykat

Kartu yoonu jaaykat yi ñooy wane man-man yi sa ekip mëna tabax ci kanam.

Njëgg ak budget

Anamu jënd ak jaay ak tànneefi dugal dañu am njeexital ci njëg ak risk ci diir bu xawa yàgg.

Risk ak kaaraange

Li liggéeyukaay bi di ñaax mooy tëral ni produit bi di doxee, kaaraange gi ak ubbeeku gi.

Ëlëgu LAION ak ay done yu ubbeeku

Done yu ubbeeku yi dina ñu jànkoonte ak fitna buy gëna yokk ci wàllu yelleefu moomeel, nangu, ak ëmbiit buy lore, di push ci segg bu gëna am doole, dajale bu xam lisence, ak registre yu baña bàyyi. LAION gennewaat benn done buñ raxas dafay wane coppite ci audit kaaraange muy jéego buñ jagleel. Xaarandi yeneen done yuñ defar wala yuñ am lisaas, seeni sàrti bopp ak jumtukaayi gis. Tension bi am ci digganté open access ngir laboratoire yu ndaw yi ak risk legal ak ethique yi ci done yuñ raxas ci web bi mooy leeral fase bi ci topp ci tabax dataset.

Doxal ci àdduna dëgg

Taggat xeetu bind-ci-nataal bu ubbeeku lu melni Diffusion bu dëgër ci kaw ay miliyaar ci ñaari nataal-kapsiyon

Tabax ak méngale sistemu seet nataal-bind bu nuru CLIP ak sistemu xaaj bu amul benn nataal

Gëstu ci wàllu done, kaaraange ëmbiit li, ak fi done yi bawoo ci eskaalu web bi

Segg ensemble yu ndaw ci làkk, dayo, wala poñ estetik ngir sos ensemble done yuñ jagleel

Risk yi ak balustrade yi

Koom-koomu ubbite mën na raw stabilite ci def liggéeyu defar dëgg.

Njëg yi ci API wala coppite ci sàrt yi mën nañu dindi xalaat yi ci guddi gi.

Dependence ci benn jaaykat dafay yokk njëgu tëjug ak migraasioŋ.

Roadmap ngir samp gi

1

Saytu sa fournisseur yi nga jëfandikoo sa liggéey ak say done.

2

Xoolaat mbir yu nëbbu, kaaraange ak sàrti yoon balaa ngay boole.

3

Fexe am palaŋu fallback ci model yi wala jaaykat yi.

4

Xool notu génne yi suko defee coppite yi ci kàrtu yoon du jaaxal ekip yi.

Weyal di banneexu

Free newsletter

Get the daily AI briefing

Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.

One email each weekday. Unsubscribe in one click. We never sell or share your address.

Test yourself

Take the LAION and Open Datasets quiz

Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.

Tambalil quiz

Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation

Laaj yi ñuy faral di laaj

What is LAION and Open Datasets?

LAION kuréel la guy def te Yàlla tax, génne ay done yu bari yu ubbeeku ci nataal, li ci gëna siiw mooy LAION-5B, moo jàppale tàggat model yu ubbeeku yu melni Stable Diffusion. Dafa am solo ndax dafa may gëstukat yi nekk ci biti liggéeyukaay yu mag yi ñu mëna am done yu bari ci eskaalu web bi.

Lan la LAION di njëkka séddale ci ay done yu am nataal-tekst?

LAION du jël ay nataal; dafay séddale ay URL ak ay metadata, suko defee jëfandikukat yi jël nataal yi ci seen balluwaay web bi njëkk.

Lu tollu ci ñaata mbind ak nataal ñoo am ci LAION-5B?

LAION-5B amna luy tollu ci 5.85 milyaari nataal-bind, moo tax ñu tuddee ko '5B'.

Ban model la LAION jëfandikoo ngir segg ñaari nataal-tekst ngir am kalite bu méngoo?

LAION dafa jëfandikoo CLIP bu OpenAI ngir natt nurug nataal-biir yi ak sànni ñaar yi méngoowul bu baax.

Ban xeetu generatif bu ubbeeku lañu tàggat ci jëfandikoo done LAION?

Stable Diffusion, xeetu bind-ci-nataal bu ubbeeku, dañu ko tàggat ci done LAION nataal-bind.

Ban jafe-jafe bu rëy la gëstukat yi gis ci LAION-5B ci 2023?

Gëstukat yi gis nañu ay lëkkalekaay ci mbirum xañ xale yudul yoon, loolu waral LAION dindi done yi, setal leen, ba noppi gennewaat beneen bu gëna wóor.