Dellu ci Passage Dense
Dense Passage Retrieval (DPR) dafay wut mbind mu am njariñ ci méngale lu laaj bi ak pàcc yi di tekki ci ay vecteur numérique, te duñu méngoo ak baat yi.
Résumé
It matters because it can retrieve correct answers even when the query and the document share zero vocabulary.
Plongeur bu xóot
DPR, bi Facebook AI dugal ci 2020, dafay jëfandikoo ñaari encodeur BERT yu wuute: encodeur laaj ak encodeur passage. Ku nekk ci ñoom dafay soppi mbind mu nekk vecteur bu dëgër bu am guddaay bu takku (dafay faral di am 768 dimension). Relevans mooy produit dot bi am ci digganté vecteur laaj ak vecteur passage, kon recuperation bi dafay nekk seetlu bu gaaw ci dëkkandoo bi gëna jege ci kaw embedding passage yuñ njëkka xayma. Modèle bi dañu ko tàggat ak mébet bu wuute: dindi vecteur bi ci passage bu ndeyjoor bi jege laaj bi te puus bu baaxul bi, jëfandikoo negatif yi ci lots boole ci negatif yu dëgër yiñ jëlee ci BM25. Ci benchmarks QA yu ubbeeku yu melni Laajte Natural, DPR moo raw BM25 bi yàgg a am doole ci marge yu bari, di wane ni méngoo semantik bi ñu jàng mën na gëna am njariñ ci seetlu baatu-caabi ngir tontu laaj yi.
Gis-gis xarala
DPR bi-encoder la: dafay encoder laaj bi ak passage bu nekk ci boppam, kon vecteur passage yépp ñu ngi leen di xayma benn yoon ba noppi denc leen ci benn index vecteur (lu melni, FAISS). Sooy laaj, dangay enkode laaj bi kese, ba noppi nga def seetug dëkkandoo bi gëna jege. Taggat yaram dafay aju ci negatif yi ci biir-batch - yeneen paas ci benn mini-batch bi dañuy nekk misaali negatif yu daanaka duñu fay, loolu dafay tax benn peer bu baax mëna defar ay méngale yu bari yu wuute ci anam wu jaar yoon.
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 jëlaat paas bu dëgër
Leegi retrieval bu dëgër mooy gëna dooleel gasoduc yiy yokk retrieval yuy dundal xeetu làkk yu bari. Gëstu baa ngi jëm ci sistem hybrid yuy boole poñ yu dëgër ak lexical, xeetu interaction yu mujj yu melni ColBERT yuy tëye vecteur per-token ngir gëna méngoo, ak tegtal yuñ defar ngir mëna ànd ak liggéey yu bari. Xaarandi encoder yu gëna yomb, yu bari làkk, ak yu gëna gudd, boole ci tàggat bu gëna dëgër diggante retrievers yi ak generatër yi ñuy liggéey.
Doxal ci àdduna dëgg
Sistemu tontub laaj bu ubbeeku biy génne ay pàccu Wikipedia yuy jàppale balaa LLM di bind tontu li
Seetug këyitu liggéeyukaay, di barab bu liggéeykat yi di laajte ci anam wu jaar yoon, ba noppi am paragraf yu am solo doonte amul baatu-caabi
Robot yiy jàppale kiliyaan yi dañuy jël xët wi war ci santu ndimbal ci plainte buñ joxe leeral
chatbots yuñ yokk ci seetlu tontu ci base de connaissance privée ngir wàññi hallucination
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 Dense Passage Retrieval 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
ColBERT Recuperation interaction tarde
Laaj yi ñuy faral di laaj
What is Dense Passage Retrieval?
Dense Passage Retrieval (DPR) dafay wut mbind mu am njariñ ci méngale lu laaj bi ak pàcc yi di tekki ci ay vecteur numérique, te duñu méngoo ak baat yi. Dafa am solo ndax mën na joxe tontu yu jaar yoon doonte laaj bi ak këyit bi bokku ñu benn vocabulaire.
Lan mooy xalaat bu mag bi ci ginaaw Recuperation Passage Dense?
DPR dafay xayma laaj yi ak paas yi ci ay vecteur yu dëgër, ba noppi natt njariñ li ci seeni nuru (produit dot), di jàpp luñuy tekki moo gën ñuy jaxasoo ci baat yi.
DPR dafay jëfandikoo 'bi-encoder'. Loolu lumuy tekki ?
Bi-encoder amna encoder yu moom seen bopp ngir laaj ak passage, suko defee ñu mëna xayma vecteur passage yi bu njëkk ba noppi indexe leen bu njëkk.
Lan moo waral ñu mëna xayma vecteur passage yi ci DPR?
Ndax encodeur passage bi nekkul ci laaj bi, bépp passage embedding mën nañu ko njëkka calculer ba noppi denc ko, bàyyi laaj bi kese ngir encoder ci waxtu seetlu bi.
Ban xeetu tàggat yaram mooy may DPR mu defar misaal yu bari yu baaxul ci njëg yu yomb?
Negatif yi ci biir lots yi dañuy jàppee yeneen passage yi ci biir lots bu ndaw ni negatif ci benn laaj buñ jox, loolu mooy génne ñaari mbir yu wuute ci anam wu jaar yoon.
Ban jumtukaay lañuy faral di jëfandikoo ngir gaaw ci seet dëkkandoo yi gëna jege ci DPR?
Index vecteur yu melni FAISS dañuy def seetlu bu gaaw ci dëkkandoo yi gëna jege ci kaw ay milioŋ ciy embeddings yuñ njëkka xayma.