Nataalu Autoregresif
Parti (Yoon Autoregressive Text-to-Image) dafay defar ay nataal ci anam wi modeli làkk yi di bindee frase yi: benn token nataal benn yoon, di wax luy topp ci lépp lu jiitu.
Résumé
It matters because it showed that simply scaling a sequence model can produce strikingly detailed, prompt-faithful images.
Plongeur bu xóot
Parti dafay jàppee defar nataal ni jafe-jafe tekki làkk ci toppalante, lu melni tekki masin. Tokenizer ViT-VQGAN dafay njëkka kode ab nataal ci ay token yu wuute yuñ jëlee ci benn téere kode buñ jàng. Encodeur Transformer dafay jàng liñu laaj ci bind bi, ba noppi decodeur Transformer dafay defar jeton nataal yi ci anam wu autorégresif, bu nekk ci ñoom mingi aju ci mbind mi ak ci jeton yiñ njëkka génne. Ginaaw ñu defaree token yépp, dekodeeru tokenizer bi dafay defaraat pixel yi. Google dafa yokk Parti ci 350 milioŋ dem ba ci 20 milyaar ciy parametre, ba noppi kalite nataal bi ak njubluwaayu mbind mi di gëna yokk bu baax ak dayo bi. Modèle 20B dafa ànd ak ay composition yu gudd, bind bu leer, ak ay detay yu rafet yuñ jox cër. Parti itam dugal na benchmark bu PartiPrompts, muy lu ëpp 1600 laaj yu jafe, yu lalu ci kategori yu bari ak niveau jafe-jafe.
Gis-gis xarala
Màndarga biy màndargaal mooy autorégression bu sell ci kaw ay token yu wuute: model bi dafay factoriser nataal bi ni produit bu am probabilite token bu ci topp, bu nuróo ci ngëm ak defar mbind ci stil GPT. Loolu dafay boole gis-gis bi ak làkk wi ci benn rëset tàggat, ba noppi may ko mu jël fukki at ci pexe yu toppalante. Njëg li mooy decodage yu toppalante, ndax dañu wara defar ay token ci yoon, loolu mooy tax defar bi gëna yeex ci anamu paralel, waaye dafay gëna yokk te dafay am njariñ ci model yu gëna mag.
njeextalu pexe
Gaawaay ak yaatuwaay
Visual IA mën na otomatise saytu, gis ak etiketu liggéey ci eskaal.
Tabax tànneef
Ekipu kreatif yi mën nañu defar konsept yu gëna gaaw te duñu def lu bari ci loxo.
Ekip ak def liggéey
Liggéeyukaay yi mën nañu jëfandikoo siñaal nataal wala wideo yu jafewoon lool ci liggéey.
Ëlëgu yoon Parti Nataal Autoregressif
Nataal autoregressif mingi am dekki ndax benn yax bi mën na modele mbind, nataal, audio, ak wideo ni benn stream token, loolu mooy tax ñu mëna modele multimodal. Gëstu mingi xeex ñakk kattan gu mag gi, jël misaal yu yeex, ak dekodaas speculatif, wax luy waaja am ci token paralel, ak tokenizers yu gëna baax. Xaarandil core autoregressive yi ci biir assistant general yiy boole jàng, xalaat, ak defar nataal, ak gis yooni scaling di push njubte composition ak rendering bind bu wóor ci nataal bi gëna sori.
Doxal ci àdduna dëgg
Rendering seen yu bari mbir yu jafee xam ci ay tegtal yu gudd, lu ci melni xeetu baayima yi, mbir yi ak foond yi.
Defar nataal yu am kàddu wala màndarga yuñ mëna jàng, fu komànd autoregresif di jàppale bind bi ci anam wu jaar yoon.
Benchmarking ak test stress sistem bind-ci-nataal jëfandikoo suite PartiPrompts ci kategori yu melni xam-xam àdduna ak konsept abstrait.
Defar ay tegtal yu leer ngir ay laaj yuy laaj lim bu leer ak lëkkaloo ci diggante mbir yu bari.
Risk yi ak balustrade yi
Yelleefi nataal ak nangu mën na nekk risku yoon sudee fi ñu bawoo leerul.
Performance model bi mën na wuute ci leeraay bi, demographie bi ak environmaa bi.
Njuumteg positive yi mën nañu dem te kenn duko seetlu fileek xool wuñu buntu wóolu sa bopp.
Roadmap ngir samp gi
Mandargal kritërium nangug njub, woowaat ak njëgu njuumte.
Saytu ak done yu méngoo ak anam yi ñuy liggéeyee dëgg.
Yokk jàngat nit ngir xam fu wóorul dara wala am njeexital yu rëy.
Toppal model drift bi nga baaxal ko ginaaw bi kamera bi wala done yi soppeekoo.
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 Parti Pathways Autoregressive Imaging quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Gis bi ci topp
Defar nataal yuy dellu ginaaw
Laaj yi ñuy faral di laaj
What is Parti Pathways Autoregressive Imaging?
Parti (Yoon Autoregressive Text-to-Image) dafay defar ay nataal ci anam wi modeli làkk yi di bindee frase yi: benn token nataal benn yoon, di wax luy topp ci lépp lu jiitu. Dafa am solo ndax dafa wane ni scaling model sequence rek mën na defar nataal yu leer te gaaw te wóor.
naka la Parti di defaree ab nataal?
Parti autoregressive la: dafay defar ay token nataal yu toppalante, bu nekk ci mbind mi ak ci token yiñ njëkka defar.
Ban kaadar la Parti di jëfandikoo ngir liggéeyum bind-ci-nataal?
Parti dafay jàppee defar ni tekki masin: encoder bi dafay jàng mbind, decoder bi dafay génne ay token nataal, muy tabb toppalante bu yàgg.
Lan la escaling Parti ba 20 milyaar ci parametre yi wane?
Bi Parti di màgg joge ci 350M dem ba ci 20B, njub ak njub ñoom ñaar dañu yokk, boole ci composition bu gëna baax ak bind bu leer.
Lan mooy soppi nataal ci ay token yu wuute ngir Parti?
Parti dafay jëfandikoo ViT-VQGAN ngir enkode nataal yi ci ay token yu wuute yu dekodeeru Transformer bi jàng xam luy am.
Lan mooy njariñu dekodage autorégresif bu Parti?
Ndax token bu nekk mingi aju ci yi ko jiitu, generation bi dafay toppalante te gëna yeex ci pexe yu paralel yi, doonte mën nañu ko seetlu.