ZML: Isibuyekezo — Ukusebenza Okuphezulu Kunoma Iyiphi i-Chip (NVIDIA, AMD, TPU Nokunye)

#Quasa #QUA #zml

I-ZML yinkampani yobuchwepheshe bengqalasizinda ye-AI ezinze eParis, eyakha uhlelo lwe-inference olunezingxenye ezisebenza ndawonye nolulungele ukusetshenziswa ezinhlelweni ezisebenzayo, oluhlukanisa imisebenzi ye-AI nezingxenye zekhompyutha ezilawulwa umkhiqizi othile. Ifilosofi yayo eyinhloko — “Ukusuka Kumodeli Kuya Ezingxenyeni Zekhompyutha” — ivumela noma iyiphi imodeli ukuthi isebenze kuzisheshisi ezihlukahlukene (NVIDIA, AMD, TPU, Trainium nokunye) kusetshenziswa isisekelo esisodwa sekhodi, kuyilapho kufinyelelwa ekusebenzeni okuphezulu kwezingxenye zekhompyutha.

Ngokungafani nezinhlaka ezivamile ezincike ezinhlelweni ezisindayo zokuqalisa ze-Python nasezingqimbeni ezifihla imininingwane yokusebenza, i-ZML ihlanganisela amamodeli ngokuqondile izingxenye zekhompyutha isebenzisa i-Zig ne-MLIR. Lokhu kunciphisa kakhulu izinsiza ezengeziwe ezidingekayo, kuthuthukise ukubikezeleka kokusebenza, kwandise isivinini futhi kwenze umsebenzi wonjiniyela wesofthiwe ube lula kakhulu. Le nkampani ikwenqaba ngokucacile ukusetshenziswa kwezimo ezifihlekile, izindlela zokufihla imininingwane ezisebenza njengomlingo, nezinhlelo zokuqalisa ezincike kakhulu ku-Python, ikhethe ukucaca, ikhono lokuhlanganisa izingxenye nokusebenza kahle.

I-ZML isanda kukhipha i-ZML/LLMD, iseva enamandla ye-inference yama-LLM, futhi isivele yaziwa emphakathini we-AI, yaze yasekelwa esidlangalaleni ngu-Yann LeCun, owawina i-Turing Award. Lolu hlelo lwezingxenye ezisebenza ndawonye lwakhelwe ngokuqondile izindawo zokusetshenziswa kwangempela lapho ukusebenza kahle nokuguquguquka kwezingxenye zekhompyutha kubaluleke kakhulu khona.

I-ZML ifaneleka kakhulu konjiniyela be-AI, emaqenjini e-ML, ezinkampanini ezisafufusa nasezinkampanini ezinkulu ezisebenzisa amamodeli ngezinga elikhulu futhi ezifuna ukugwema ukuboshelwa kumhlinzeki oyedwa, kuyilapho zifinyelela esivinini esiphezulu kakhulu se-inference esingatholakala kuma-chip ahlukahlukene.

Okugqamile

• I-inference engancikile ngempela ezingxenyeni zekhompyutha ezithile enesisekelo esisodwa sekhodi.
• Ukuhlanganiselwa ngokuqondile izisheshisi ezihlukahlukene ukuze kufinyelelwe ekusebenzeni okuphezulu.
• Izinsiza ezengeziwe ezidingekayo zincane kakhulu futhi onjiniyela besofthiwe basebenza kalula nangempumelelo.
• Ukugxila kakhulu ekulungele ukusetshenziswa ezinhlelweni ezisebenzayo nasekubikezelekeni kokusebenza.

Okungase Kudingeke Kucatshangelwe

• Njengoba kuyisixazululo esisasha nesinobuchwepheshe obujulile, amaqembu ajwayele izinhlaka ezigxile ku-Python angase adinge isikhathi sokufunda ukusisebenzisa.
• Sifaneleka kakhulu kubasebenzisi ababeka phambili amandla okusebenza nokulawula.

Isinqumo Esiphelele: Izinkanyezi ezingu-4.5/5

I-ZML igqama njengenye yezindlela ezinemigomo ephakeme kakhulu nezicophelela kakhulu kwezobuchwepheshe ze-inference ye-AI ngo-2026. Ngokuhamba “isuka kumodeli iye ezingxenyeni zekhompyutha,” le nkampani ixazulula izinkinga zangempela zokusetshenziswa ezinhlelweni ezisebenzayo eziphathelene nokusebenza kahle, ukuguquguquka nokwaneliseka konjiniyela besofthiwe. Iyisixazululo esithembisayo kunoma ubani ozimisele ngokusebenzisa i-AI esebenza kahle kakhulu futhi ekwazi ukusebenzisa izingxenye zekhompyutha ezihlukahlukene.

Nawe thola umklomelo we-QUA ngokubhala isibuyekezo ku-Quasa.io!

Qala: https://quasa.io/projects/zml

#Quasa #QUA #zml #QuasaRewards