Liquid AI Ihlolwa: Ikusasa le-AI Esebenza Ngokuqondileyo Kumadivayisi Ngale Kwama-Transformer

#Quasa #QUA #liquidai

I-Liquid AI (evezwe ku-Quasa.io/projects/liquid-ai) yinkampani eyasungulwa isuka ku-MIT, eyakha isizukulwane esitsha se-Liquid Foundation Models (LFMs) ezisebenza ngobuciko kungamoshwa izinsiza — izakhiwo ezingasebenzisi ama-transformer, eziklanywe kusukela ezimisweni eziyisisekelo ukuze zisetshenziswe emhlabeni wangempela.

Esikhundleni sokugijimisa amamodeli asemafini aqhubeka esiba makhulu, i-Liquid igxile ku-AI esebenza kuhle kakhulu, elesikhathi esifitshane sokulinda impendulo, njalo esebenza ngokuqondileyo kumadivayisi: amafoni, amakhompyutha aphathwayo, izimota, amadivayisi agqokwayo, amarobhothi, lemishini esebenzela lapho idatha ekhiqizwa khona.

Okuyisisekelo se-Liquid AI yikuletha ubuhlakani obukhulu kakhulu kusetshenziswa amandla amancane kakhulu okubala, ngokuklama amamodeli kucatshangelwa imishini azasebenza kuyo.

Okumqoka okugoqelwayo:

• Liquid Foundation Models (LFM2 series) — Ukusebenza okusezingeni eliphezulu kakhulu kusetshenziswa inkumbulo lamandla kagesi okuncane kakhulu. Amamodeli asukela ku-350M kusiya ku-24B+ wamapharamitha, kugoqela izinhlobo ze-Mixture-of-Experts ezivula ingxenye encane kuphela yamapharamitha ku-token ngayinye.
• Ukusebenza Kudivayisi & Ngokuqondileyo Lapho Idatha Ekhona — Isikhathi sokulinda impendulo esingaphansi kwe-20ms, ama-ejenti abiza amathuluzi, amamodeli ahlanganisa ukubona lolimi (LFM2-VL), amamodeli omsindo, lamamodeli okucabanga asebenza ngokupheleleyo kungelakho ukuxhumana le-inthanethi, kugcinwa idatha iyimfihlo ngokupheleleyo.
• LEAP Platform & Liquid Apollo — Amathuluzi alula kubathuthukisi okulungiselela ama-LFM imisebenzi ethile, ukuwalolonga ngokuqondileyo, lokuwafaka ukuze asebenze kuma-CPU, kuma-GPU lakuma-NPU.
• Izixazululo Zamabhizimusi & Lezimboni — Amamodeli enzelwe izidingo ezithile embonini yezimota (Mercedes-Benz), ekuthengiseni nge-inthanethi, kwezezimali, kwezemithi (kusetshenzwa le-Insilico Medicine), lasebuchwephesheni bamarobhothi.
• Ukufakwa Ukuze Kusebenze Ngendlela Ehlanganisiweyo & Eyimfihlo — Sebenzisa i-AI elamandla emafini, emishinini esendaweni yenhlanganiso, kumbe ngokupheleleyo kudivayisi — ilungele kakhulu izimboni ezilawulwa yimithetho lezinhlelo zokusebenza ezidinga impendulo masinyane.

Ilungele kakhulu onjiniyela be-AI, amaqembu athuthukisa imikhiqizo, abenzi bezimota, izinkampani zamarobhothi, izinkampani zemikhiqizo yama-elekthronikhi yabathengi, amabhizimusi asakhulayo akha i-AI elezimpawu eziyehlukanisayo, layo yonke inhlanganiso edinga ubuhlakani obuphanga ukusebenza, obugcina idatha iyimfihlo, njalo obungabizi kakhulu lapho idatha ekhiqizwa khona. Ngo-2026, i-Liquid AI ikhokhela ukuguquka kusukwa ku-AI esebenza emafini kuphela kusiya ebuhlakani obutholakala ngeqiniso yonke indawo, obusebenza ngokuqondileyo kumadivayisi.

Umphakathi wabathuthukisi lowamabhizimusi uyayithanda:
“Ekucineni, amamodeli ayisisekelo asebenza ngeqiniso kukhompyutha yami ephathwayo ngesiqubu esisebenzisekayo, kungelazindleko zamafu.”
“Ama-LFM e-Liquid enze kangcono kakhulu kulalokho ebesikudinga ngesikhathi sokulinda impendulo ku-AI yangaphakathi kwemota, egcina ledatha iyimfihlo ngokupheleleyo.”
“Amamodeli la azwakala eyikusasa — asebenzisa izinsiza ngobuciko, aphatha izinhlobo ezitshiyeneyo zedatha, njalo angafakwa ngempela ukuze asebenze loba kungaphi.”

Isebenza kuhle kakhulu ikakhulu ekusebenziseni i-AI kudivayisi ukuze kukhutshwe imiphumela, ezinhlelweni zokusebenza ezibeka ukugcinwa kwemfihlo kuqala, emarobhothini asebenzisa amandla amancane lasemisebenzini yezimota, ekulungiseleleni amamodeli imisebenzi ethile masinyane, lasekuletheni ukusebenza okuphezulu kungelamaqoqo amakhulu kakhulu ama-GPU.

Ubuthakathaka: Isesakhiwo esitsha nxa iqathaniswa lama-transformer abusa kakhulu (uhlelo lwamathuluzi lezinsiza eziyisekelayo lukhula masinyane kodwa kalukavuthwa ngezinga elifanayo); imiphumela emihle kakhulu ivame ukuvela ekuthuthukiseni ukusebenza kuqondiswe emishinini ethile; umsebenzi owenzelwa izidingo zamabhizimusi yindlela ephumelela kakhulu emisebenzini eyinkimbinkimbi.

Sekukonke, kumaqembu ngo-2026 afuna i-AI elamandla ecabangela imikhawulo yasemhlabeni wangempela yesikhathi sokulinda impendulo, amandla kagesi, ukugcinwa kwemfihlo, lemishini, i-Liquid AI ingenye yezinkampani zamamodeli ayisisekelo ezithokozisa kakhulu njalo eziletha izixazululo ezisebenzisekayo. Kayisomunye nje umhlinzeki we-LLM — iguqula indlela i-AI esebenza ngayo lendawo etholakala kuyo emhlabeni wangempela.

Zuza lomvuzo we-QUA nge-Quasa!

Izinkanyezi ezingu-4.9/5 (ivelele ekusebenziseni izinsiza ngobuciko, ekusebenzeni kumadivayisi angempela, lasesakhiweni esicabangela ikusasa; uhlelo lwezinsiza eziyisekelayo luyakhula).

Qalisa: https://quasa.io/projects/liquid-ai