Liquid AI: Uphononongo lwekamva le-AI esebenza ngqo kwizixhobo, ngaphaya kwee-transformer

#Quasa #QUA #liquidai

I-Liquid AI (eboniswe ku-Quasa.io/projects/liquid-ai) yinkampani ephuma kwi-MIT eyakha isizukulwana esitsha see-Liquid Foundation Models (LFMs) ezisebenza ngobuchule — izakhiwo ezingasebenzisi ii-transformer, eziyilwe kusetyenziswa imigaqo esisiseko ukuze zisetyenziswe kwiimeko zokwenene.

Endaweni yokusukela iimodeli zelifu eziya zisanda ngobukhulu, i-Liquid igxile kwi-AI esebenza ngamandla, enokulibaziseka okuncinane nesebenza ngqo kwizixhobo: iifowuni, iilaptop, iimoto, izixhobo ezinxitywayo, iirobhothi nezixhobo zekhompyutha ezikufutshane nalapho idatha iveliswa khona.

Eyona nto isisiseko kwi-Liquid AI kukubonelela ngobukrelekrele obukhulu kusetyenziswa amandla amancinane ekhompyutha, ngokuyila iimodeli kuthathelwa ingqalelo izixhobo eziza kusebenza kuzo.

Iimpawu eziphambili ziquka:

• Ii-Liquid Foundation Models (uthotho lwe-LFM2) — Ukusebenza okukwinqanaba eliphezulu lobuchwepheshe bangoku, kusetyenziswa imemori namandla amancinane kakhulu. Iimodeli zisusela kwiipharamitha ezingama-350M ukuya kwezingama-24B+, kuquka iintlobo ze-Mixture-of-Experts ezenza kusebenze iqhezu kuphela leepharamitha ngethokheni nganye.
• Ukusebenza Kwisixhobo & Ngokuthe Ngqo Apho Idatha Iveliswa Khona — Ukulibaziseka okungaphantsi kwe-20ms, ii-arhente ezibiza izixhobo, iimodeli zombono nolwimi (LFM2-VL), iimodeli zesandi neemodeli zokuqiqa ezisebenza ngokupheleleyo ngaphandle kweintanethi, zigcina ubumfihlo ngokupheleleyo.
• I-LEAP Platform & ne-Liquid Apollo — Izixhobo ezilula zabaphuhlisi zokwenza ii-LFMs zilungele imisebenzi ethile, ukuzilungelelanisa kakuhle nokuzisebenzisa kwii-CPUs, kwii-GPUs nakwii-NPUs.
• Izisombululo Zamashishini & Namacandelo Oshishino — Iimodeli ezenzelwe iimfuno ezithile kushishino lweemoto (Mercedes-Benz), kurhwebo lweintanethi, kwezemali, kushishino lwamayeza (kunye ne-Insilico Medicine), nakwezobuchwepheshe beerobhothi.
• Ukusetyenziswa Okuxubeneyo & Nokwabucala — Sebenzisa i-AI enamandla efini, kwiziseko zenkampani, okanye ngokupheleleyo kwisixhobo — ilungele kakhulu amacandelo oshishino alawulwa yimimiselo nee-app apho ukulibaziseka kufuneka kube kuncinane kakhulu.

Ifaneleke kakhulu kwiinjineli ze-AI, kumaqela emveliso, kubavelisi beemoto, kwiinkampani zeerobhothi, kwiibrendi zezixhobo ze-elektroniki zabathengi, kwiinkampani ezisakhulayo ezakha i-AI eneempawu ezahlukileyo, nakuwo nawuphi na umbutho ofuna ubukrelekrele obukhawulezayo, obugcina ubumfihlo nobungabizi kakhulu, kufutshane nalapho idatha iveliswa khona. Ngo-2026, i-Liquid AI ikhokela inguqu esuka kwi-AI esebenza efini kuphela iye kubukrelekrele obufumaneka ngokwenene kuyo yonke indawo nobusebenza ngqo kwizixhobo.

Uluntu lwabaphuhlisi nolwamashishini luyayithanda kakhulu:
“Ekugqibeleni, iimodeli ezisisiseko ezisebenza ngokwenene kwilaptop yam ngesantya esisebenzisekayo, kungekho zindleko zelifu.”
“Ii-LFMs ze-Liquid zigqithe kakhulu kwiimfuno zethu zokunciphisa ukulibaziseka kwe-AI esemotweni, ngoxa zigcina idatha iyimfihlo ngokupheleleyo.”
“Ezi modeli zivakala ngathi zezekamva — zisebenzisa izibonelelo ngobuchule, zisebenza ngeentlobo ezininzi zedatha, yaye zinokusetyenziswa ngokwenene naphi na.”

Ibalasele ngakumbi ekusebenziseni i-AI kwisixhobo ukwenza uqikelelo nezigqibo, kwii-app ezibeka ubumfihlo phambili, kwiirobhothi ezisebenzisa amandla amancinane nakwiimeko zokusetyenziswa ezinxulumene neemoto, ekwenzeni iimodeli zilungele imisebenzi ethile ngokukhawuleza, nasekuboneleleni ngokusebenza okukwinqanaba eliphezulu ngaphandle kwamaziko amakhulu azele zii-GPU.

Iintsilelo: Isesisakhiwo esitsha xa sithelekiswa nee-transformer ezixhaphakileyo (inkqubo exhasayo nezixhobo zikhula ngokukhawuleza kodwa azikavuthwa ngokulinganayo); iziphumo ezizezona zilungileyo zidla ngokuvela ekuphuculeni ukusebenza kulungiselelwa izixhobo ezithile; umsebenzi owenzelwe iimfuno ezithile zamashishini yeyona ndlela inamandla kwiimeko zokusetyenziswa ezintsonkothileyo.

Lilonke, kumaqela ango-2026 afuna i-AI enobuchule ethathela ingqalelo imida yokwenene yokulibaziseka, yamandla, yobumfihlo neyezixhobo zekhompyutha, i-Liquid AI yenye yeenkampani zeemodeli ezisisiseko ezichulumancisa kakhulu nezisebenziseka kakuhle. Ayisosiboneleli se-LLM esongezelelweyo nje — ichaza ngokutsha indlela nendawo i-AI ekhoyo nesebenza kuyo ngokwenene.

Fumana nomvuzo we-QUA nge-Quasa!

Iinkwenkwezi ezi-4.9/5 (ibalasele ekusebenziseni izibonelelo ngobuchule, ekusebenzeni kwizixhobo zokwenene nakwisakhiwo esijonge kwikamva; inkqubo exhasayo iyakhula).

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