Jagoran Harshe AI

Cross-Encoders vs Bi-Encoders

Hanyoyi biyu na ƙirar jijiyoyi suna kwatanta rubutu: masu ɓoye biyu suna haɗa kowane yanki daban don bincike mai sauri, yayin da masu rikodin giciye suna karanta duka rubutun tare don daidaito mafi girma.

2 min karatuAn sabunta ta ƙarshe

Dubawa

The choice shapes the speed-versus-precision tradeoff in every modern search and retrieval system.

Zurfafa nutsewa

Dukansu gine-ginen suna amsa 'yaya alaƙar rubutu biyu suke?', amma sun bambanta a lokacin da nassin suka hadu. Mai rikodin bi-encoder yana tafiyar da kowace jimla ta hanyar transfoma da kansa, yana samar da tsayayyen vector guda ɗaya akan kowane rubutu; kamanni sannan samfurin ɗigo ne mai arha ko cosine tsakanin vectors. Saboda ana iya ƙididdige ƙwayoyin cuta a gaba da adana su, masu rikodin bi-encoders sun kai miliyoyin takardu da ma'ajin bayanai masu ƙarfi. Mai rikodin giciye a maimakon haka yana haɗa rubutun ([CLS] tambaya [SEP] daftarin aiki) kuma yana ciyar da su ta hanyar ƙira tare, barin kowane alama ya halarci kowane alamar kafin fitar da madaidaicin maki guda ɗaya. Wannan cikakkiyar kulawa tana ɗaukar ingantacciyar ma'amala mai rikodin rikodi ta ɓace, don haka masu rikodin giciye sun fi daidai amma ba za su iya ƙididdige komai ba kuma dole ne su gudana sau ɗaya a kowane biyu.

Fahimtar Fasaha

Babban bambanci shine iyakar kulawa. A cikin rikodi bi-biyu, hankalin kai baya ketare tsakanin abubuwan shiga guda biyu, don haka abubuwan da aka saka daftarin aiki sun zama masu zaman kansu na tambaya kuma ana iya sake amfani da su. A cikin mai rikodin giciye, hankali yana ɗaukar jerin abubuwan da aka haɗa, yana mai da alamar tambaya ta dogara. Ma'auni na farashi daidai da haka: Takaddun martaba N suna buƙatar N cikakken taswirar wucewa don mai rikodin giciye tare da kwatancen vector mai arha don mai rikodi guda biyu bayan lambar tambaya ɗaya.

Dabarun Tasiri

Gudu da sikelin

Gudun aikin harshe na iya tafiya da sauri ba tare da sadaukar da daidaito ba.

Shiga ku isa

Yana faɗaɗa damar shiga cikin harsuna da salon sadarwa.

Shawarwari masu haske

Ƙungiyoyi za su iya ciyar da ƙarin lokaci akan hukunci yayin da aiki da kai ke sarrafa maimaitawa.

Makomar Cross-Encoders vs Bi-Encoders

Mafi rinjayen tsarin shine maidowa-sa'an nan-sake-sake: mai-encoder yana debo 'yan takara ɗari kaɗan daga miliyoyin, sa'an nan mai rikodin giciye ya sake yin odar babban sakamako. Samfurin ma'amala na ƙarshe kamar ColBERT sun raba bambance-bambance ta hanyar adana kayan aikin kowane-token, kuma distillation yana ƙara horar da ƙaramin encoders don yin koyi da hukunce-hukuncen giciye. Yi tsammanin masu gyara masu rahusa da haɗin kai na matakai biyu zuwa bututun da aka haɓaka.

Aiwatar da Gaskiyar Duniya

Ma'ajin bayanai na vector yana amfani da abubuwan da aka haɗa bi-encoder don dawo da manyan hanyoyin 200 na ɗan takara daga miliyoyin takardu a cikin millise seconds.

Mai rikodin rikodin ketare yana sake ba da odar waɗancan 'yan takara 200 kafin a ciyar da su zuwa RAG chatbot, inganta ingantaccen amsa.

Jumla-Transformers jiragen ruwa waɗanda aka riga aka horar da su biyu-encoders (don bincike na ma'ana) da maɓallan giciye (don ƙarawa da maki STS)

Gano kwafin-tambaya akan dandalin Q&A yana amfani da mai rikodin giciye don madaidaicin madaidaicin hanya biyu akan jerin zaɓi

Hatsari & Tsare-tsare

Abubuwan da aka ruɗe suna iya shigar da rahotanni cikin nutsuwa, kwararar tallafi, ko abubuwan bincike.

Hankali na gaggawa na iya ƙirƙirar sakamako mara daidaituwa a cikin buƙatun iri ɗaya.

Za a iya fallasa bayanan rubutu mai ma'ana idan ikon samun dama yana da rauni.

Taswirar Hanya

1

Ƙayyade tsarin fitarwa, sautin, da ma'auni masu inganci kafin fitowa.

2

Amsa a ƙasa tare da amintattun tushe a duk lokacin da daidaito ya shafi mahimmanci.

3

Ajiye wurin binciken ɗan adam don abubuwan da ake samu masu girma.

4

Bibiyar tsarin gazawar kuma sake horar da tsokaci ko tafiyar aiki akai-akai.

Ci gaba da Bincike

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-Encoders vs Bi-Encoders quiz

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

Fara tambayoyi

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

Jagora na gaba

BERT da Model Encoder

Tambayoyin da ake yawan yi

What is Cross-Encoders vs Bi-Encoders?

Hanyoyi biyu na ƙirar jijiyoyi suna kwatanta rubutu: masu ɓoye biyu suna haɗa kowane yanki daban don bincike mai sauri, yayin da masu rikodin giciye suna karanta duka rubutun tare don daidaito mafi girma. Zaɓin yana siffanta saurin-da-madaidaicin ciniki a cikin kowane tsarin bincike na zamani.

Menene ma'anar bambance-bambancen tsarin gine-gine tsakanin mai rikodin giciye da mai rikodi?

Ƙirar-ƙira-ƙira ta haɗa duk abubuwan da aka shigar kuma bari hankali ya ta'allaka duka jerin; Bi-encoders suna ɓoye kowane rubutu a keɓance cikin ɓoyayyiyar ɓarna.

Me yasa masu rikodin bi-encode suka daidaita zuwa miliyoyin takardu yadda ya kamata?

Domin kowace takarda tana ƙunshe da kanta, ana iya sake amfani da vector ɗinta a duk tambayoyin kuma ana iya ƙididdige su kafin lokaci.

A cikin bututun bincike na yau da kullun, ta yaya ake haɗa gine-ginen biyu?

Madaidaicin tsarin dawo da-sannan-sake-sake yana amfani da mai rikodin rikodi mai sauri don faɗuwar tunowa da ingantacciyar mai rikodin giciye don sake tsara jerin sunayen.

Me yasa maɓalli na giciye ba zai iya ƙididdige wakilcin daftarin aiki ba?

Tunda an samar da makin daga jerin takaddun da aka haɗa, ya dogara da tambaya kuma dole ne a ƙididdige shi ga kowane nau'i biyu.

Menene maɓalli guda biyu ke fitarwa don rubutu guda ɗaya?

Mai rikodin rikodi guda biyu yana tsara kowane rubutu zuwa vector guda ɗaya; sannan ana lissafta kamance tsakanin vectors.