Fàttaliku bu jàkkarloo
Cross-attention mooy anam wi ñuy may benn sekans mu xool beneen: dekodeer biy defar mbind mën na bàyyi xel ci ni enkodeer bi di wane liñ dugal.
Résumé
It is how models connect what they are producing to what they read, powering translation, captioning, and modern multimodal systems.
Plongeur bu xóot
Self-attention dafay tax token yi ci benn rang ñu mëna déngoo seen biir; cross-attention dafay tax benn toppalante jëlee xibaar ci beneen bu wuute. Ci biir decodeur Transformer, jéego bu nekk ci defar yi dafay forme ay laaj ci génnekaay biñ defaree ci wàll, fekk caabi yi ak valeur yi ci génnekaayu encodeur bi lañuy bawoo. Modèle bi dafay xayma njariñu élément input bu nekk ci position output bi fi nekk, ba noppi di génne njaxas bu pondéré ci input yi. Loolu mooy tax dekodeeru tekki làkk mëna bàyyi xelam ci baatu cosaan bi war, bi muy bind baat bu nekk. Lu weesu bind, cross-attention mooy kol bi ci model multimodal yi: dekodeeru bind mën na bàyyi xel ci màndarga patch nataal yi, wala model audio mën na méngale son bi ak kàddu yiñ bind. Saa yoo bëggee boole ñaari xeeti xibaar yu wuute, cross-attention mooy faral di nekk tisu biy boole.
Gis-gis xarala
Ci anam wu mekanik, bàyyi xel ci cross-attention dafay jëfandikoowaat benn formul produit dot buñ scaleer bi ak bàyyi xel ci sa bopp, ak benn coppite: laaj yi dañuy joge ci benn rang (decodeur bi) ba noppi caabi/valeur yi dañuy joge ci beneen rang (encodeur bi). Dafay xayma poid yiy bàyyi xel ni softmax ci kaw nuru butoŋu laaj, ba noppi delloo limu valeur yuñ pondéree. Ndax laaj yi ak caabi yi ci balluwaay yu wuute lañu bawoo, ñaari toppalante yi mën nañu wuute ci guddaay, ci anam, wala ci làkk lëmm.
njeextalu pexe
Gaawaay ak yaatuwaay
Liggéeyukaay yi ci làkk yi mën nañu gëna gaaw te duñu yàq deggoo gi.
Dugg ak yegg
Dafay yaatal jëfandikoo gi ci làkk yi ak ci anam yi ñuy jokkoo.
dogal yu gëna leer
Ekip yi mën nañu gëna yàgg ci àtte ci jamono ji otomatisation di liggéey ci baamtu.
Ëlëgu bàyyi xel ci jàmm
Cross-attention mooy gëna bari interface buñ miin ngir boole xeeti tombi ñaw yi. Modèlu làkku gis-gis dañu koy jëfandikoo ngir mbind mëna dëkk ci gox yi am nataal; defarkatu nataalu diffusion dañu koy jëfandikoo ngir defar pixel yi ci laaj mbind. Gëstu yi dañuy puus ci cross-attention bu gëna am njariñ (variante yu ligneer ak yu néew) ngir mëna jëfandikoo dokimaa yu gudd, nataal yu am resolusioŋ bu kawe, ak wideo. Ginaaw sistemu IA yi dañuy boole ay yëg-yëg yu bari, xaarnaa ni ay lalu cross-attention ñu nekk ay lëkkalekaay yuy boole mbind, son, gis-gis ak ay done yuñ yamale.
Doxal ci àdduna dëgg
Ci tekki làkku masin neuronal, dekodeer bi dafay xoolaat baat yiñ jëlee ngir tànn tekki bu dëggu bi ci baat bu nekk.
Diffusion bu dëgër dafay jëfandikoo cross-attention ngir mëna tànn gox bu nekk ci nataal yiñ defar ci laajtu bind bi.
Modèlu làkku gis-gis bu melni Flamingo dafay may token mbind mu jàkkarloo ak man-mani nataal yi ngir tontu laaj yi ci gis-gis.
Dekodeer yiy soppi kàddu-ci-tekst dañuy boole kadre audio yiñ kode ngir méngale son yi ak kàddu yi ñuy bind.
Risk yi ak balustrade yi
Lépp lu jaarul yoon mën na dugg ci rapoor yi, jàppale ci liggéey bi, wala ci njariñu gëstu bi.
Sensibilite bu gaaw mën na jur njariñ yu wuute ci laajte yu noonu mel.
Done yu am solo mën nañu feeñ sudee seytu jëfandikoo gi néew doole.
Roadmap ngir samp gi
Mandargal formaa génne gi, melokaan bi, ak standard kalite yi laata ngay dugal ko.
Tontu yu am solo ak balluwaay yu wóor saa yu dëggu bi di am solo.
Fexeel am barabu xool nit ñi ngir am njariñ yu am solo.
Toppal anami gacce yi ak di faral di tàggataat ay laaj wala def-liggéey.
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 Cross-Attention 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
Soppi-Fàttaliku-Xaajal Soppi
Laaj yi ñuy faral di laaj
What is Cross-Attention?
Cross-attention mooy anam wi ñuy may benn sekans mu xool beneen: dekodeer biy defar mbind mën na bàyyi xel ci ni enkodeer bi di wane liñ dugal. Mooy anam wi model yi di boolee li ñuy defar ak li ñuy jàng, tekki làkk, kaption ak sistem multimodal yu bees yi.
Lan mooy wuutale bàyyi xel ci sa bopp ak bàyyi xel ci yeneen mbir?
Cross-attention dafay xëcc laaj yi ci benn rang (lu melni, decoder bi) ak caabi ak valeur yu bawoo ci beneen (lu melni, encoder bi), bàyyi ñaar ñi ñu jëflante.
Ci biir Transformateur encodeur-decodeur, fan la laajte yi jëm ci cross-attention di jogee?
Dekodeer bi dafay formu ay laaj ci génnekaay bimu defar ci wàll, ba noppi jëfandikoo génnug enkodeer bi ni caabi ak valeur ngir génne leeral yi war.
Lan moo waral ñaari toppalante yi ci cross-attention mën nañu am guddaay wala anam yu wuute?
Ginaaw laajte yi ci benn balluwaay lañuy bawoo, caabi yi/valeur yi ci beneen balluwaay, toppalante yi mën nañu wuute ci guddaay, làkk, wala anam.
naka lay def ba diffusion stable di jëfandikoo cross-attention?
Cross-attention dafay may wàll wu nekk ci nataal biñ defar mu bàyyi xel ci bind bi, daal di gëdd nataal yi ci kàddu yi.
Ban jëf ci math la cross-attention ak bàyyi sa bopp bokk?
Ñoom ñaar ñépp ñu ngi jëfandikoo benn produit dot-produit scale: poñ yu nuróo diggante laaj ak caabi, softmax, ginaaw ga benn somme de valeur yuñ pondéree.