Sabbin Samfura
Reranker samfuri ne na mataki na biyu wanda ke sake ƙididdige jerin sunayen sakamakon bincike don dacewa da tambaya, yana ƙara yin oda bayan mai dawo da sauri ya ja 'yan takara.
Dubawa
It is a key ingredient in modern search and retrieval-augmented generation (RAG).
Zurfafa nutsewa
Bincike da tsarin RAG yawanci suna aiki a matakai biyu. Na farko, mai saurin dawo da sauri (sau da yawa binciken bincike / haɗawa ko mahimmin kalmar BM25) yana jan ƙila takaddun ɗan takara 50-100 daga miliyoyi - an inganta su don tunawa da sauri. Amma wannan wucewa ta farko yana ƙididdige tambayoyin da takaddun daban, don haka zai iya rasa kuskure. Mai sake fasalin shine ainihin matakin: yana ɗaukar tambaya da kowane ɗan takara tare kuma ya fitar da madaidaicin mahimmin ma'auni, sannan ya sake tsara lissafin don haka mafi kyawun sakamako ya tashi zuwa sama. Babban tsarin gine-ginen shine maɓallin ketare: yana ciyar da tambaya da daftarin aiki tare zuwa na'urar wuta, yana barin kowace alamar tambaya ta halarci kowace alamar takarda. Wannan zurfafan hulɗar yana sa masu sakewa su zama daidai fiye da haɗa kamanni, a farashin gudu sau ɗaya ga kowane ɗan takara.
Fahimtar Fasaha
Bambance-bambancen shine bi-encoder tare da mai rikodin giciye. Mai rikodin rikodi guda biyu yana haɗa tambaya da daftarin aiki daban-daban zuwa ɓangarori, don haka kamanni samfurin ɗigo ne mai arha - mai sauri kuma wanda za'a iya lissafinsa, amma mara zurfi. Mai rikodin giciye yana haɗa tambaya da daftarin aiki zuwa shigarwa ɗaya kuma yana gudanar da cikakken izinin canza canji, yana samar da ma'auni guda ɗaya mai dacewa tare da kyakkyawar kulawar matakin alama. Ba za a iya ƙididdige shi ba, don haka an keɓe shi don sake tsara ƙaramin jerin sunayen da aka zaɓa. Samfura kamar Cohere Rerank da BGE-reranker sun misalta wannan.
Dabarun Tasiri
Kudin da kasafin kuɗi
Hukunce-hukuncen gine-gine suna haifar da aiki da tsadar aiki na shekaru.
Shawarwari masu haske
Ilimin fasaha yana taimaka wa ƙungiyoyi su zaɓi tari mai kyau, ba kawai sabon abu ba.
Kula da inganci
Zaɓuɓɓukan injiniya mafi kyau suna rage abin dogaro a cikin samarwa.
Makomar Sabunta Model
Rerankers suna zama daidaitattun a cikin bututun RAG saboda ingantaccen tsari mahallin yana inganta ingancin amsa LLM kai tsaye kuma yana rage hasashe. Yi tsammanin mafi sauƙi, masu rikodin giciye masu sauri, masu jujjuya harsuna da yawa da multimodal (rubutu da hotuna ko teburi), da tsayin mahallin windows don haka za a iya zira kwallaye gabaɗaya. Madogarar 'jeri' na tushen LLM waɗanda ke yanke hukunci ga ɗan takarar da aka saita lokaci ɗaya suna girma, kuma wasu tsarin suna karkatar da hukunce-hukuncen ƙetaren giciye zuwa masu dawo da rahusa don samun daidaito kusa da matakin farko.
Aiwatar da Gaskiyar Duniya
RAG chatbot yana maidowa 50 chunks ta hanyar shigar da bincike, sa'an nan kuma sake canza ra'ayi don ciyar da manyan 5 mafi mahimmanci kawai cikin mahallin LLM.
Binciken kasuwancin e-kasuwanci yana sake yin odar sakamakon samfur don haka abubuwan da suka fi dacewa da cikakken jumlar tambayar mai siyayya su bayyana a farko
Cohere Rerank ko BGE-reranker yana haɓaka madaidaicin binciken daftarin aiki akan dubban manufofin PDFs.
Tushen ilimin tallafin abokin ciniki yana maido da abubuwan taimako da aka samo don haka wakilin ya fito da amsar da ta fi dacewa a saman.
Hatsari & Tsare-tsare
Haɓaka ma'auni ɗaya na iya ɓoye manyan raunin tsarin.
Sau da yawa ana raina kayan more rayuwa da kuma kuɗin kulawa.
Tsaro da gibin lura na iya girma yayin da tsarin ke ƙara haɓaka.
Taswirar Hanya
Ƙayyade latency, inganci, da maƙasudin farashi kafin aiwatarwa.
Alamar ma'auni a ƙarƙashin ainihin kaya da yanayin bayanai.
Kula da kayan aiki don kurakurai, ɗigo, da tasirin mai amfani.
Shirya bijirowa da hanyoyin mayar da martani kafin sikeli.
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 Reranking Models 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
Jagora na gaba
Gwajin A/B don Samfuran ML
Tambayoyin da ake yawan yi
What is Reranking Models?
Reranker samfuri ne na mataki na biyu wanda ke sake ƙididdige jerin sunayen sakamakon bincike don dacewa da tambaya, yana ƙara yin oda bayan mai dawo da sauri ya ja 'yan takara. Yana da maɓalli mai mahimmanci a cikin bincike na zamani da haɓaka-ƙaramar haɓaka (RAG).
A cikin bututun mai da matakai biyu na al'ada, menene aikin reranker?
Mai saurin dawo da 'yan takara; mai reranker sannan ya sake saka wannan jerin sunayen don tura mafi dacewa sakamakon zuwa sama.
Wane gine-ginen da akasarin masu gyara ke amfani da shi?
Rerankers yawanci suna amfani da maɓallan giciye, tambayar ciyarwa da daftarin aiki tare don haka hankali zai iya ƙirƙira mu'amalarsu.
Me yasa ba za a iya amfani da reranker mai rikodin giciye don bincika miliyoyin takardu kai tsaye ba?
Ba kamar abubuwan haɗawa da za a iya ƙididdige su ba, mai rikodin giciye dole ne ya gudanar da cikakken taswirar fasfo kowane biyu, don haka an tanada shi don ƙaramin jerin sunayen.
Menene babban fa'idar mai-encoder akan mai rikodin giciye?
Bi-encoders sun haɗa takaddun da kansu kuma a gaba, suna ba da damar binciken kamanni cikin sauri a cikin tarin tarin yawa.
Ta yaya masu yin rerankers ke taimakawa rage hasashe a cikin tsarin RAG?
Mafi kyawun tsari, mahallin da ya fi dacewa yana ba LLM ingantaccen ƙasa, wanda ke rage ƙirƙira amsoshin.