Aggrégation njaxasu Agent
Mixture-of-Agents (MoA) xarala la boo xamni xeeti làkk yu bari dañuy bind ay tontu, ginaaw ga ñu boole seeni xalaat yu gëna baax ci benn tontu bu gëna baax.
Résumé
It lets a team of open models rival or beat a single top-tier model.
Plongeur bu xóot
Dugalug ci benn këyitu 2024 bu bawoo ci Together AI, Njaxasu-Agents dafay amaale LLM yu bari ci ay diisaay. Ci layer bu njëkk bi, barina xeetu 'proposer' yu ku nekk ci boppam tontu laaj bi. Ginaaw loolu ñu boole seeni génnekaay, daal di koy jox layer bi ci topp, foofu la model yi di tontuwaat, leegi ñu ngi aju ci bépp projet bu njëkk. Ginaaw benn wala ñaari yoon yu mel noonu, xeetu 'agrégateur' bu mujj dafay boole lépp ci benn tontu. Li gëna am solo ci gis-gis bi, bi bindkat yi di woowe 'liggéeyandoo ci LLM yi', mooy xeetu tontu yi dañuy defar tontu yu gëna baax suñu leen wanee tontu seeni moroom, doonte tontu yu jaarul yoon. Ci AlpacaEval 2.0 benchmark, benn MoA buñ tabax lëmm ci xeetu open-source lañu wax ni dafa raw poñ yi GPT-4 Omni am, loolu dafay wane ni dajale bu baax xeetu model yu bari te yomb mën na raw benn sistem frontiere.
Gis-gis xarala
MoA wuute na ak wote bu yomb: du tann benn tontu, aggregator dafay jàng tontub kàndidaa yépp ni contexte ba noppi defar ab synthese bu bees, jaxase doole ak njuumte segg. Diversité ci biir proposer yi dafay jàppale, kon jaxase ay famiy model yu wuute lu am solo la. Jariñ bi dafa am ay layer, melni reso bu xóot fu layer bu nekk 'neuron' nekk woote LLM lëmm. Kompromis bi mooy latency ak njëg: layer bu nekk dafay yokk limu woote inference yi, moo tax MoA dafay dugal xaalis bu bari ci calcul ngir yéegal kalite bi.
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 njaxasu aggregation
Xaarandi agrégation bu nuroo ak MoA ngir tasaaroo ndax inference bi dafay gëna xéewale ak kaadar orchestration bi gëna màgg. Gëstu yi ci yoon yi bokkuna ci jàng ban proposer lañu wara wóolu ci laaj bu nekk (routing), wàññi latency penalty ci daw proposer yi ci parallel ak dagg ñi néew doole teel, ak boole MoA ak jumtukaay-jëfandikoo agents suko defee aggregator fuses du bind rek waaye jëf ak firnde yuñ jëlee. Lu model yu ubbeeku yi di gëna bari, boole leen ci anam wu am xel nekk na yoon wu gëna am solo ngir yegg ci niveau frontière bu amul benn model bu mag.
Doxal ci àdduna dëgg
Boole ñetti xeeti chat yu ubbeeku yu wuute ni ay proposer, ba noppi jëfandikoo ab agrégateur bu dëgër ngir defar benn tontu buy jàppale kiliyaan yi.
Yokkatal poñ yi topp tegtal yi ci xeetu AlpacaEval, di jëfandikoo xeetu open-source kese.
Boole xalaati kod yu bari yu bawoo ci model yu bari ci benn jëfandikoo fonction bu gëna am doole.
Doxal ab pipeline bu ubbeeku buy jegesi kalite frontiere ngir ab dugal bu xarañ ci wàllu kumpa, fu done yi mënu génn ci serveur yu benn kompiñi.
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 Mixture-of-Agents Aggregation 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
Njaxasu xóotaayu
Laaj yi ñuy faral di laaj
What is Mixture-of-Agents Aggregation?
Mixture-of-Agents (MoA) xarala la boo xamni xeeti làkk yu bari dañuy bind ay tontu, ginaaw ga ñu boole seeni xalaat yu gëna baax ci benn tontu bu gëna baax. Dafay may ekipu model yu ubbeeku yi ñu xëcc wala ñu daan benn model bu gëna kawe.
Ci Njaxasu-Agent yi, lan mooy wareefu xeetu aggregatër bi?
Agrégateur bi dafay jàng tontub kàndidaa yépp ba noppi defar benn tontu bu boole te gëna baax.
Ban moomeel bu am solo ci LLMs la këyitu MoA sukkandikoo ci?
Auteur yi dañu gis ni model yi dañuy gëna baaxal seen tontu suñu sukkandikoo ci yeneen model yu ñu bind, doonte yu jaarul yoon.
Lan mooy wuutale MoA ak wote yu bari yu yomb?
MoA wuute na ak wote, MoA du tann benn tontu rek; Dafay defar tontu bu bees buy boole kàndidaa yi.
Lan moo waral wuute ci xeetu xalaat yi am solo ci MoA?
Xeeti proposer yu bari dañuy muur suuf su bari, loolu dina tax aggregator bi am xalaat yu bari yu mu wara boole ak njuumte yu mu wara segg.
Lan mooy njëgu jëfu MoA bi gëna mag?
Layer bu nekk dafay yokk yeneen woote LLM, kon MoA dafay jënd yeneen xayma ak diir ngir gëna baaxal tontu.