https://quasa.io/zu/video/prime-intellect-qeqesha-i-ai-esezingeni-eliphezulu-ngaphandle-kwezinkampani-ezinkulu-zobuchwepheshe-incazelo-ephelele 2026-11-10T23:30:26+00:00 https://i.ytimg.com/vi/On47OM1RTZc/hqdefault.jpg Prime Intellect: Qeqesha i-AI Esezingeni Eliphezulu Ngaphandle Kwezinkampani Ezinkulu Zobuchwepheshe #Quasa #QUA #PrimeIntellect I-Prime Intellect yakha i-Open Superintelligence Stack — inkundla ehlanganisa zonke izingxenye zobuchwepheshe evumela noma ubani ukuba aqeqeshe, afake ekusebenzeni futhi aqhubeke ethuthukisa amamodeli akhe e-AI ngaphandle kokuthembela engqalasizinda yezinkampani ezinkulu zobuchwepheshe. Inhloso yenkampani ukwenza i-AI esezingeni eliphezulu kakhulu ifinyeleleke futhi ibe ngeyabanikazi ngokuhlanganyela ngokusebenzisa amandla okucubungula angalawulwa yisikhungo esisodwa, amathuluzi anekhodi yomthombo evulekile kanye nentuthuko eholwa umphakathi. Emnyombweni wayo, i-Prime Intellect inikeza indawo yokusebenza edidiyelwe ehlanganisa amandla okucubungula, ukuqeqesha ngokufunda ngokuqiniswa (RL), ukuhlola, ukusebenzisa amamodeli ukukhiqiza imiphumela kanye nezindawo zokuhlola ezihlukanisiwe neziphephile. Abasebenzisi bangafinyelela izinkulungwane zezindawo zokusebenza ze-RL ezakhiwe umphakathi nge-Environment Hub, benze ukuqeqesha ngezinga elikhulu ezingqalasizindeni ezisingathiwe, bafake amamodeli ekusebenzeni ngokusekelwa kwe-LoRA okwakhelwe ngaphakathi, futhi basebenzise imakethe yomhlaba wonke yama-GPU esukela kumanodi ngamanye iye kumaqoqo amakhulu kakhulu. Inkampani isizuze impumelelo ehlaba umxhwele kwezobuchwepheshe. Ikhiphe i-INTELLECT-2, imodeli yokuqala enamapharamitha angu-32B eqeqeshwe ngokufunda ngokuqiniswa okusatshalaliswe emhlabeni wonke, yalandelwa yi-INTELLECT-3 (imodeli ye-Mixture-of-Experts enamapharamitha angu-100B+) kanye ne-SYNTHETIC-2 (izigidi zemikhondo yokucabanga ekhiqizwe ngokubambisana). Imitapo yayo yekhodi yomthombo evulekile — i-Verifiers ne-Prime-RL — yenza kube nokwenzeka ukusebenzisa i-RL esatshalalisiwe ngezinga elikhulu, enezinqubo ezingadingi ukulindana futhi ekwazi ukuqhubeka nokusebenza lapho kuba namaphutha. Isekelwa abatshalizimali abahamba phambili nabantu abanombono wekusasa le-AI, okuhlanganisa i-Founders Fund, i-Radical Ventures, i-NVIDIA, i-Intel, u-Andrej Karpathy, u-John Schulman nabanye, i-Prime Intellect ihlanganisa ucwaningo olusezingeni eli https://www.youtube.com/embed/On47OM1RTZc 2026-11-10T23:30:26+00:00 https://quasa.io/xh/video/prime-intellect-qeqesha-i-ai-ekwinqanaba-eliphambili-ngaphandle-kwe-big-tech-ingcaciso-epheleleyo 2026-11-10T23:30:55+00:00 https://i.ytimg.com/vi/On47OM1RTZc/hqdefault.jpg Prime Intellect: Qeqesha i-AI Ekwinqanaba Eliphambili Ngaphandle kwe-Big Tech (Ingcaciso Epheleleyo) #Quasa #QUA #PrimeIntellect I-Prime Intellect yakha i-Open Superintelligence Stack — iqonga eliquka zonke iinkalo zobugcisa elivumela nabani na ukuba aqeqeshe, asebenzise aze aphucule ngokuqhubekayo iimodeli zakhe ze-AI ngaphandle kokuxhomekeka kwiziseko zobugcisa ze-Big Tech. Injongo yale nkampani kukwenza i-AI ekwinqanaba eliphambili ifikeleleke kwaye ibe yeyabantu ngokudibeneyo ngokusebenzisa amandla okubala angalawulwa liziko elinye, izixhobo ezinekhowudi evulelekileyo, nophuhliso olukhokelwa luluntu. Undoqo wenkonzo ye-Prime Intellect yimeko yokusebenza edibeneyo equka amandla okubala, uqeqesho lokufunda ngokuqiniswa (RL), uvavanyo, ukuveliswa kweziphumo ngeemodeli, neendawo zovavanyo ezikhuselekileyo nezizimeleyo. Abasebenzisi banokufikelela kumawaka eemeko ze-RL ezakhiwe luluntu nge-Environment Hub, baqhube uqeqesho olubanjwe eqongeni ngomlinganiselo omkhulu, basebenzise iimodeli ngenkxaso eyakhelwe ngaphakathi ye-LoRA, baze basebenzise indawo yentengiso yehlabathi yee-GPU, ukususela kwiindawo zokubala ezizimeleyo ukuya kumaqela amakhulu kakhulu azo. Le nkampani sele ifezekise impumelelo yobugcisa encomekayo. Ikhuphe i-INTELLECT-2, imodeli yokuqala eneeparamitha eziyi-32B eqeqeshwe ngokufunda ngokuqiniswa okusasazwe kwihlabathi liphela, yalandelwa yi-INTELLECT-3 (imodeli ye-Mixture-of-Experts eneeparamitha eziyi-100B+) ne-SYNTHETIC-2 (izigidi zemizila yokuqiqa eveliswe ngentsebenziswano). Amathala ayo ekhowudi evulelekileyo — i-Verifiers ne-Prime-RL — avumela i-RL esasaziweyo, engadingi kungqamaniswa nengamelana neempazamo, ngomlinganiselo omkhulu. Ixhaswa ngabatyali-mali abaphambili nabantu abanombono wekamva le-AI, kuquka i-Founders Fund, i-Radical Ventures, i-NVIDIA, i-Intel, u-Andrej Karpathy, u-John Schulman, nabanye, i-Prime Intellect idibanisa uphando olukumgangatho ophezulu neziseko zobugcisa ezisebenzisekayo. Abathengi abafana ne-Ramp ne-Zapier sele besebenzisa eli qonga ukwakha iiarhente ezikhethekileyo eziqhuba ngcono kuneemodeli ezikwinqanaba eliphambili kwimisebenzi ethile. I-Prime Intellect i https://www.youtube.com/embed/On47OM1RTZc 2026-11-10T23:30:55+00:00 https://quasa.io/sn/video/prime-intellect-dzidzisa-ai-yepamusoro-soro-pasina-big-tech-tsananguro-yakazara 2026-11-10T23:30:56+00:00 https://i.ytimg.com/vi/On47OM1RTZc/hqdefault.jpg Prime Intellect: Dzidzisa AI Yepamusoro-soro Pasina Big Tech (Tsananguro Yakazara) #Quasa #QUA #PrimeIntellect Prime Intellect iri kuvaka Open Superintelligence Stack — puratifomu ine zvikamu zvose zvinodiwa inoita kuti chero munhu akwanise kudzidzisa, kuisa mukushanda, uye kuramba achivandudza mamodheru ake eAI asingavimbi nezvivakwa zvetekinoroji zveBig Tech. Chinangwa chekambani iyi ndechekuita kuti AI yepamusoro-soro iwanikwe uye ive yevanhu pamwe chete kuburikidza nesimba remakomputa risina kuunganidzwa panzvimbo imwe chete, maturusi ane kodhi yakavhurika, uye kuvandudzwa kunotungamirirwa nenharaunda. Pakati pezvainopa, Prime Intellect ine nzvimbo yakabatanidzwa inosanganisa simba remakomputa, kudzidzisa kuburikidza nekudzidza kunosimbiswa nemibayiro (RL), ongororo, kushandisa mamodheru kupa mhinduro, uye nzvimbo dzakaparadzaniswa dzakachengeteka dzekumhanyisa zvirongwa. Vashandisi vanogona kuwana zviuru zvenzvimbo dzeRL dzakavakwa nenharaunda kuburikidza neEnvironment Hub, kuita kudzidzisa pamwero mukuru pazvivakwa zvinopiwa nepuratifomu, kuisa mamodheru mukushanda aine rutsigiro rweLoRA rwakavakirwa mukati, uye kushandisa musika wepasi rose wemaGPU anotangira pamakomputa anoshanda ari oga kusvika kumapoka makuru kwazvo emakomputa akabatana. Kambani iyi yakatoita zvinoyemurika munyaya dzetekinoroji. Yakaburitsa INTELLECT-2, modheru yekutanga ine maparamita 32B yakadzidziswa kuburikidza nekudzidza kunosimbiswa nemibayiro kwakaparadzirwa pasi rose, ichiteverwa neINTELLECT-3 (modheru yeMixture-of-Experts ine maparamita 100B+) neSYNTHETIC-2 (mamiriyoni enhevedzano dzematanho ekufunga dzakagadzirwa pamwe chete). Maraibhurari avo ane kodhi yakavhurika — Verifiers nePrime-RL — anoita kuti RL yakaparadzirwa ikwanise kuitwa pamwero mukuru, zvikamu zvayo zvichishanda zvisingamirirani uye ichiramba ichishanda kunyange pakakanganisika zvimwe zvikamu. Ichitsigirwa nevanoisa mari vanotungamira uye vane chiono cheramangwana reAI vanosanganisira Founders Fund, Radical Ventures, NVIDIA, Intel, Andrej Karpathy, John Schulman, nevamwe, Prime Intellect inosanganisa tsvakurudzo yepamusoro-soro nezvivakwa zvet https://www.youtube.com/embed/On47OM1RTZc 2026-11-10T23:30:56+00:00 https://quasa.io/ny/video/prime-intellect-phunzitsani-ai-yapamwamba-kwambiri-popanda-kudalira-makampani-akuluakulu-aukadaulo-kufotokoza-mwatsatanetsatane 2026-11-10T23:31:13+00:00 https://i.ytimg.com/vi/On47OM1RTZc/hqdefault.jpg Prime Intellect: Phunzitsani AI Yapamwamba Kwambiri Popanda Kudalira Makampani Akuluakulu Aukadaulo #Quasa #QUA #PrimeIntellect Prime Intellect ikumanga Open Superintelligence Stack — nsanja yokhala ndi zida zonse zofunika kuti aliyense athe kuphunzitsa, kuyamba kugwiritsa ntchito ndi kupitiriza kukonza ma model ake a AI popanda kudalira zomangamanga zamakampani akuluakulu aukadaulo. Cholinga cha kampaniyi n'chakuti AI yapamwamba kwambiri izipezeka kwa aliyense ndipo izikhala ya anthu onse pamodzi, pogwiritsa ntchito mphamvu zowerengera zogawidwa m'malo osiyanasiyana, zida za open-source ndi chitukuko chotsogoleredwa ndi anthu ogwiritsa ntchito nsanjayi. Maziko a Prime Intellect ndi malo ophatikiza mphamvu zowerengera, kuphunzitsa pogwiritsa ntchito njira yophunzirira kudzera mu mphotho (RL), kuyesa ma model, kugwiritsa ntchito ma model popereka mayankho ndi malo otetezeka oyeseramo mapulogalamu. Ogwiritsa ntchito angapeze malo masauzande ophunzitsiramo RL opangidwa ndi anthu ogwiritsa ntchito nsanjayi kudzera mu Environment Hub, kuphunzitsa ma model pamlingo waukulu pa makina osamalidwa ndi nsanjayi, kuyamba kugwiritsa ntchito ma model okhala ndi chithandizo cha LoRA chomangidwa kale, ndiponso kupeza ma GPU pamsika wapadziko lonse, kuyambira pa makina amodzi mpaka pamagulu akuluakulu a makina olumikizidwa pamodzi. Kampaniyi yakwaniritsa kale zinthu zochititsa chidwi pankhani yaukadaulo. Inatulutsa INTELLECT-2, model yoyamba yokhala ndi ma parameter 32B yophunzitsidwa kudzera mu RL yogawidwa pamakina padziko lonse, kenako INTELLECT-3 (model ya Mixture-of-Experts yokhala ndi ma parameter 100B+) ndi SYNTHETIC-2 (njira zoganizira mamiliyoni opangidwa mogwirizana). Malaibulale awo a open-source — Verifiers ndi Prime-RL — amathandiza kuti RL yogawidwa pamakina ambiri ichitike pamlingo waukulu, makina akugwira ntchito popanda kudikirirana ndiponso kupitiriza kugwira ntchito ngakhale ena atasokonekera. Mothandizidwa ndi osunga ndalama otsogola komanso anthu a masomphenya pa nkhani ya AI, kuphatikizapo Founders Fund, Radical Ventures, NVIDIA, Intel, Andrej Karpathy, John Schulman ndi ena, Prime Intellect imaphatikiza k https://www.youtube.com/embed/On47OM1RTZc 2026-11-10T23:31:13+00:00 https://quasa.io/rw/video/prime-intellect-toza-ai-yo-ku-rwego-rwo-hejuru-utishingikirije-ku-bigo-bikomeye-byikoranabuhanga-isesengura-rirambuye 2026-11-10T23:31:27+00:00 https://i.ytimg.com/vi/On47OM1RTZc/hqdefault.jpg Prime Intellect: Toza AI yo ku rwego rwo hejuru utishingikirije ku bigo bikomeye by'ikoranabuhanga ( #Quasa #QUA #PrimeIntellect Prime Intellect irimo kubaka Open Superintelligence Stack — urubuga rukomatanyije inzego zose rutuma buri wese ashobora gutoza, gushyira mu bikorwa no gukomeza kunoza moderi ze bwite z'ubwenge bw'ubukorano (AI) atishingikirije ku bikorwa remezo by'ibigo bikomeye by'ikoranabuhanga. Intego y'iki kigo ni ugutuma AI yo ku rwego rwo hejuru igera kuri bose kandi ikaba umutungo basangiye, binyuze mu bushobozi bwo kubara butagengwa n'ikigo kimwe, ibikoresho bifite kode ifunguye n'iterambere riyobowe n'umuryango w'ababikoresha n'ababiteza imbere. Ishingiro rya Prime Intellect ni uburyo bukomatanyije buhuza ubushobozi bwo kubara, gutoza hakoreshejwe kwiga hifashishijwe ibihembo (RL), isuzuma, gukoresha moderi mu gutanga ibisubizo n'ahantu hizewe ho kugeragereza porogaramu. Abakoresha bashobora kugera ku bihumbi by'ahantu ho gutoza hakoreshejwe RL hubatswe n'umuryango w'ababikoresha n'ababiteza imbere binyuze muri Environment Hub, gukora imyitozo ku rugero runini ku bikorwa remezo bibacumbikira, gushyira moderi mu bikorwa hakoreshejwe ubufasha bwa LoRA bwubatswemo, no gukoresha isoko mpuzamahanga rya GPU kuva ku ipfundo rimwe ry'urusobe kugeza ku matsinda manini cyane y'amapfundo. Iki kigo kimaze kugera ku bikorwa bitangaje mu bya tekiniki. Cyasohoye INTELLECT-2, moderi ya mbere ifite ibipimo 32B yatojwe hakoreshejwe kwiga hifashishijwe ibihembo byakwirakwijwe ku isi hose, ikurikirwa na INTELLECT-3 (moderi ya Mixture-of-Experts ifite ibipimo 100B+) na SYNTHETIC-2 (za miriyoni z'uruhererekane rw'intambwe zo gutekereza rwakozwe ku bufatanye). Amasomero yabo ya kode ifunguye — Verifiers na Prime-RL — atuma bishoboka gukoresha RL ikwirakwijwe ku rugero runini, aho ibikorwa bidategerezanya kandi sisitemu igakomeza gukora n'iyo habaye amakosa. Prime Intellect, ishyigikiwe n'abashoramari bakomeye n'abafite icyerekezo cy'ejo hazaza ha AI barimo Founders Fund, Radical Ventures, NVIDIA, Intel, Andrej Karpathy, John Schulman n'abandi, ihuza ubushakashatsi bugezweho cyane n'ibikorwa remezo bikoreshwa mu biko https://www.youtube.com/embed/On47OM1RTZc 2026-11-10T23:31:27+00:00 https://quasa.io/rn/video/prime-intellect-menyereza-ai-yo-ku-rwego-rwo-hejuru-udashingiye-kuri-big-tech-insiguro-yuzuye 2026-11-10T23:31:51+00:00 https://i.ytimg.com/vi/On47OM1RTZc/hqdefault.jpg Prime Intellect: Menyereza AI yo ku rwego rwo hejuru udashingiye kuri Big Tech (Insiguro yuzuye) #Quasa #QUA #PrimeIntellect Prime Intellect iriko irubaka Open Superintelligence Stack — urubuga rurimwo ibikoresho vyose bikenewe rutuma umuntu uwo ari we wese ashobora kumenyereza, gushira mu ngiro no kubandanya kunagura imideli yiwe ya AI atarinze kwisunga ibikorwa remezo vya Big Tech. Intumbero y'iryo shirahamwe ni ugutuma AI yo ku rwego rwo hejuru ishikirwa na bose kandi ikaba iy'abayifiseko uruhara bose, biciye mu bushobozi bwo guharura busabagiye, ibikoresho bifise kode yuguruye n'iterambere riyoborwa n'abagize umuryango wayo. Mu ntimatima yayo, Prime Intellect itanga uburyo bukomatanyije ubushobozi bwo guharura, ukumenyereza biciye mu kwiga gushimangira (RL), amasuzuma, ugukoresha imideli kugira itange inyishu n'ibibanza bikingiwe vyo kugeragerezamwo. Abayikoresha barashobora gushikira ibihumbi vy'ibibanza vya RL vyubatswe n'abagize umuryango biciye kuri Environment Hub, gukora imyimenyerezo yagutse ku bikorwa remezo bibakira, gushira imideli mu ngiro ifise ubufasha bwa LoRA bwubatswemwo, no gukoresha isoko ryo kw'isi yose rya GPUs, kuva ku mahuriro amwe gushika ku migwi mininiminini yayo. Iryo shirahamwe riramaze gushika ku ntambwe zikomeye mu vy'ubuhinga. Ryarasohoye INTELLECT-2, umudeli wa mbere ufise ibipimo 32B wamenyerejwe biciye mu kwiga gushimangira gusabagiye kw'isi yose, ukurikigwa na INTELLECT-3 (umudeli wa Mixture-of-Experts wa 100B+) na SYNTHETIC-2 (amamiriyoni y'uruhererekane rw'intambwe zo kuzirikana zakozwe mu bufatanye). Amasomero yabo ya kode yuguruye — Verifiers na Prime-RL — atuma bishoboka gukora RL yagutse kandi isabagiye, ibice vyayo bigakora bitarinze kurindirana kandi ikabandanya gukora n'igihe hari ibice binaniwe. Ifashwe mu mugongo n'abashoramari bakomeye hamwe n'abafise intumbero ya kure muri AI, barimwo Founders Fund, Radical Ventures, NVIDIA, Intel, Andrej Karpathy, John Schulman n'abandi, Prime Intellect ikomatanya ubushakashatsi buhanitse n'ibikorwa remezo bishobora gukoreshwa mu bikorwa. Abakiriya nka Ramp na Zapier baramaze gutangura gukoresha urwo rubuga mu kwubaka abako https://www.youtube.com/embed/On47OM1RTZc 2026-11-10T23:31:51+00:00 https://quasa.io/ln/video/prime-intellect-pesa-ai-ya-nivo-ya-likolo-mateya-kozanga-kotalela-big-tech-ndimbola-mobimba 2026-11-10T23:31:59+00:00 https://i.ytimg.com/vi/On47OM1RTZc/hqdefault.jpg Prime Intellect: Pesa AI ya nivo ya likolo mateya kozanga kotalela Big Tech (Ndimbola mobimba) #Quasa #QUA #PrimeIntellect Prime Intellect ezali kotonga Open Superintelligence Stack — plateforme oyo esangisi biteni nyonso mpe epesaka moto nyonso nzela ya kopesa ba-modele na ye ya AI mateya, kotia yango na mosala mpe kokoba kobongisa yango kozanga kotalela biloko ya Big Tech ya kosalisa yango esala. Mokano ya kompanyi ezali ya kosala ete AI ya nivo ya likolo ezala na maboko ya bato nyonso mpe ezala eloko na bango ya lisanga na nzela ya nguya ya kosala mitango oyo ekabolami kozanga esika moko ya bokonzi, bisaleli oyo code na yango ezali polele mpe bokolisi oyo lisanga etambwisaka. Na moboko na yango, Prime Intellect epesaka esika ya mosala oyo esangisi nguya ya kosala mitango, mateya na nzela ya mbano (RL), bomeki ya makoki, bopesi biyano mpe bisika ya komeka oyo ebatelami mpe ekabwani na système mosusu. Basaleli bakoki kozwa bankoto ya bisika ya mateya ya RL oyo lisanga etongi na nzela ya Environment Hub, kopesa mateya na bonene monene na ba-serveur oyo plateforme eyambaka, kotia ba-modele na mosala na lisungi ya LoRA oyo ezali kati na yango mpe kosalela zando ya mokili mobimba ya ba-GPU, banda na ba-noeud moko moko tii na bituluku minene mpenza. Kompanyi esili kokokisa makambo ya tekiniki ya kokamwa. Ebimisaki INTELLECT-2, modele ya liboso ya 32B ba-parametre oyo ezwaki mateya na nzela ya RL oyo ekabolami na mokili mobimba, na nsima INTELLECT-3 (modele ya Mixture-of-Experts ya 100B+) mpe SYNTHETIC-2 (bamilio ya milongo ya matambe ya kokanisa oyo esalemaki na boyokani). Babiblioteke na bango oyo code na yango ezali polele — Verifiers mpe Prime-RL — epesaka nzela ya kosala RL na bonene monene na ba-machine ebele, oyo esalaka na bantango ekeseni mpe ekobaka kosala ata soki biteni mosusu ebebi. Na lisungi ya batyami ya mosolo ya liboso mpe bato oyo bazali na emoneli ya mosika mpo na AI, kati na bango Founders Fund, Radical Ventures, NVIDIA, Intel, Andrej Karpathy, John Schulman mpe basusu, Prime Intellect esangisaka bolukiluki ya nivo ya likolo mpenza na biloko ya tekiniki oyo ekoki kosalelama na misala ya sol https://www.youtube.com/embed/On47OM1RTZc 2026-11-10T23:31:59+00:00 https://quasa.io/tn/video/prime-intellect-katisa-ai-ya-maemo-a-a-kwa-godimo-thata-ntle-le-big-tech-tlhaloso-ka-botlalo 2026-11-10T23:32:14+00:00 https://i.ytimg.com/vi/On47OM1RTZc/hqdefault.jpg Prime Intellect: Katisa AI ya Maemo a a Kwa Godimo Thata Ntle le Big Tech (Tlhaloso ka Botlalo) #Quasa #QUA #PrimeIntellect Prime Intellect e aga Open Superintelligence Stack — polatefomo e e akaretsang dikarolo tsotlhe e e letlang mongwe le mongwe go katisa, go tsenya mo tirisong le go tswelela a tokafatsa dimmotlele tsa gagwe tsa AI ntle le go ikaega ka mafaratlhatlha a Big Tech. Maikaelelo a khamphani ke go dira gore AI ya maemo a a kwa godimo thata e fitlhelelwe ke botlhe le go nna ya botlhe ka bokgoni jwa khomphiutha jo bo sa laolweng go tswa mo lefelong le le lengwe, didirisiwa tsa khoutu e e bulegileng le tlhabololo e e eteletsweng pele ke setšhaba. Mo motheong wa yone, Prime Intellect e fana ka tikologo e e kopantsweng e e akaretsang bokgoni jwa khomphiutha, katiso ya go ithuta ka tuelo (RL), ditlhatlhobo, go dirisa dimmotlele go ntsha dipholo le ditikologo tsa diteko tse di sireletsegileng. Badirisi ba ka fitlhelela diketekete tsa ditikologo tsa RL tse di agilweng ke setšhaba ka Environment Hub, ba tsamaisa katiso e kgolo e e amogelwang mo mafaratlhatlheng a polatefomo, ba tsenya dimmotlele mo tirisong ka tshegetso ya LoRA e e ageletsweng mo polatefomong, mme ba dirisa mmaraka wa lefatshe lotlhe wa di-GPU, go simolola ka dinouto ka bongwe go fitlha ka ditlhopha tse dikgolo thata tsa tsone. Khamphani e setse e fitlheletse dilo tse di kgatlhang mo go tsa thekenoloji. E ntshitse INTELLECT-2, mmotlele wa ntlha o o nang le dipharametara di le 32B o o katisitsweng ka go ithuta ka tuelo mo go anameng le lefatshe lotlhe, wa latelwa ke INTELLECT-3 (mmotlele wa Mixture-of-Experts wa 100B+) le SYNTHETIC-2 (dimilione tsa metlhala ya go akanya e e tlhagisitsweng ka tirisanommogo). Dilaeborari tsa bone tsa khoutu e e bulegileng — Verifiers le Prime-RL — di kgontsha RL e e anameng ka selekanyo se segolo, e dikarolo tsa yone di sa tlhokeng go dira ka nako e le nngwe e bile e tswelela e dira le fa dikarolo dingwe di ka palelwa. Ka tshegetso ya babeeletsi ba ba eteletseng pele le batho ba ba bonelang pele isagwe ya AI, go akaretsa Founders Fund, Radical Ventures, NVIDIA, Intel, Andrej Karpathy, John Schulman le ba b https://www.youtube.com/embed/On47OM1RTZc 2026-11-10T23:32:14+00:00 https://quasa.io/st/video/prime-intellect-kwetlisa-ai-ya-boemo-bo-hodimo-ntle-le-big-tech-tlhaloso-e-felletseng 2026-11-10T23:32:22+00:00 https://i.ytimg.com/vi/On47OM1RTZc/hqdefault.jpg Prime Intellect: Kwetlisa AI ya Boemo bo Hodimo Ntle le Big Tech (Tlhaloso e Felletseng) #Quasa #QUA #PrimeIntellect Prime Intellect e haha Open Superintelligence Stack — sethala se akaretsang dikarolo tsohle se dumellang mang kapa mang ho kwetlisa, ho kenya tshebetsong le ho tswela pele ho ntlafatsa mehlala ya hae ya AI ntle le ho itshetleha ka meralo ya motheo ya Big Tech. Morero wa khamphani ke ho etsa hore AI ya boemo bo hodimo ka ho fetisisa e fihlellehe mme e be ya bohle ka kopanelo, ka matla a dikhomphutha a sa laolweng sebakeng se le seng, disebediswa tsa khoutu e bulehileng le ntshetsopele e etelletsweng pele ke setjhaba. Motheong wa yona, Prime Intellect e fana ka tikoloho e kopaneng e akaretsang matla a dikhomphutha, kwetliso ya ho ithuta ka matlafatso (RL), ditlhahlobo, ho hlahisa diphetho ka mehlala le ditikoloho tse sireletsehileng tsa teko. Basebedisi ba ka fihlella ditikoloho tse dikete tsa RL tse hahilweng ke setjhaba ka Environment Hub, ba tsamaisa kwetliso e kgolo e tshwarelwang sethaleng, ba kenya mehlala tshebetsong ka tshehetso ya LoRA e hahelletsweng kahare, mme ba sebedisa mmaraka wa lefatshe wa di-GPU ho tloha ho dinoutu tse le nngwe ho ya ho dihlopha tse kgolohadi tsa dikhomphutha. Khamphani e se e fihletse diphihlello tse tsotehang tsa botekgeniki. E lokolotse INTELLECT-2, mohlala wa pele wa dipharamitha tse 32B o kwetlisitsweng ka ho ithuta ka matlafatso ho abuweng lefatsheng lohle, o latetsweng ke INTELLECT-3 (mohlala wa 100B+ wa Motswako wa Ditsebi) le SYNTHETIC-2 (dimilione tsa tatellano ya mehato ya ho beha mabaka e hlahisitsweng ka tshebedisano). Dilaeborari tsa bona tsa khoutu e bulehileng — Verifiers le Prime-RL — di nolofalletsa RL e abuweng ka tekanyo e kgolo, e sebetsang ntle le ho emelana mme e tswellang pele le ha ho eba le diphoso. Ka tshehetso ya batsetedi ba etellang pele le ba nang le tjhebelopele ya AI, ho kenyeletswa Founders Fund, Radical Ventures, NVIDIA, Intel, Andrej Karpathy, John Schulman le ba bang, Prime Intellect e kopanya dipatlisiso tse tswetseng pele ka ho fetisisa le meralo ya motheo e sebetsang. Bareki ba kang Ramp le Zapier ba se ba sebedis https://www.youtube.com/embed/On47OM1RTZc 2026-11-10T23:32:22+00:00 https://quasa.io/ve/video/prime-intellect-gudisa-ai-ya-vhuimo-ha-ntha-u-sa-ditika-nga-big-tech-thalutshedzo-yo-fhelelaho 2026-11-10T23:32:59+00:00 https://i.ytimg.com/vi/On47OM1RTZc/hqdefault.jpg Prime Intellect: Gudisa AI ya Vhuimo ha Nṱha U Sa Ḓitika nga Big Tech (Ṱhalutshedzo yo Fhelelaho) #Quasa #QUA #PrimeIntellect Prime Intellect i khou fhaṱa Open Superintelligence Stack — pulatifomo i re na zwipiḓa zwoṱhe zwine zwa tendela muthu muṅwe na muṅwe a tshi gudisa, a tshi thoma u shumisa na u dzulela u khwinisa mimodelo yawe ya AI a sa ḓitiki nga themamveledziso ya Big Tech. Ndivho ya khamphani ndi u ita uri AI ya vhuimo ha nṱha i swikelelee na u vha thundu ya vhoṱhe nga u shumisa maanḓa a u vhalela a sa langwi fhethu huthihi, zwishumiswa zwa khoudu ya tshiko yo vuleaho na mveledziso i rangwaho phanḓa nga tshitshavha. Tshithu tsha ndeme kha Prime Intellect ndi vhupo ho ṱanganyiswaho vhune ha katela maanḓa a u vhalela, u gudisa nga u guda nga u khwaṱhisedzwa (RL), u linga, u shumisa mimodelo u bveledza phindulo na vhupo ha u lingedza ho tsireledzeaho. Vhashumisi vha nga swikelela zwigidi zwa vhupo ha RL ho fhaṱwaho nga tshitshavha nga Environment Hub, vha gudisa mimodelo nga tshikalo tshihulwane kha tshumelo dzo ṱanganedzwaho, vha thoma u shumisa mimodelo i tikedzaho LoRA yo fhaṱelwaho ngomu, vha dovha vha shumisa makete wa ḽifhasi wa dzi-GPU u bva kha node nthihi u ya kha zwigwada zwihulwane zwa khomphyutha. Khamphani yo no swikelela mvelelo dzi mangadzaho dza thekhinolodzhi. Yo bvisa INTELLECT-2, ine ya vha modeli ya u thoma i re na pharamitha dza 32B yo gudiswaho nga u guda nga u khwaṱhisedzwa ho phaḓaladziwaho ḽifhasini ḽoṱhe, ya tevhelwa nga INTELLECT-3 (modeli ya 100B+ ya Mixture-of-Experts) na SYNTHETIC-2 (mamilioni a mutevhe wa maga a u humbula o bveledzwaho nga tshumisano). Dziḽaiburari dzavho dza khoudu ya tshiko yo vuleaho — Verifiers na Prime-RL — dzi ita uri RL yo phaḓaladziwaho i kone u shumiswa nga tshikalo tshihulwane, zwipiḓa zwayo zwi tshi shuma zwi sa lindelani nahone i tshi bvela phanḓa naho hu na vhukhakhi. I tshi tikedzwa nga vhabindudzi vha rangaho phanḓa na vha re na mabono a vhumatshelo ha AI, hu tshi katelwa Founders Fund, Radical Ventures, NVIDIA, Intel, Andrej Karpathy, John Schulman na vhaṅwe, Prime Intellect i ṱanganya ṱhoḓisiso ya vhuimo ha nṱha na themamveledziso i shumi https://www.youtube.com/embed/On47OM1RTZc 2026-11-10T23:32:59+00:00 https://quasa.io/sw/video/zml-tathmini-utendaji-wa-kiwango-cha-juu-zaidi-kwenye-chipu-yoyote-nvidia-amd-tpu-na-nyinginezo 2026-11-10T23:32:59+00:00 https://i.ytimg.com/vi/Oguct9Y5QK8/hqdefault.jpg ZML Tathmini: Utendaji wa Kiwango cha Juu Zaidi kwenye Chipu Yoyote (NVIDIA, AMD, TPU na Nyinginezo) #Quasa #QUA #zml ZML ni kampuni ya miundombinu ya AI yenye makao yake Paris inayojenga mfumo wa inferensi ulio tayari kutumika katika mazingira halisi ya uzalishaji, unaotenganisha kazi za AI na maunzi yanayomilikiwa na mtoa huduma fulani. Falsafa yao kuu — “Kutoka Modeli hadi Maunzi” — huwezesha modeli yoyote kufanya kazi kwenye vichapuzi mbalimbali (NVIDIA, AMD, TPU, Trainium na vingine) kwa kutumia msingi mmoja wa msimbo huku ikifikia utendaji wa juu zaidi wa maunzi. Tofauti na mifumo ya jadi inayotegemea mazingira mazito ya kuendeshea Python na tabaka zinazoficha undani wa utekelezaji, ZML hukusanya msimbo wa modeli moja kwa moja kwa ajili ya maunzi kwa kutumia Zig na MLIR. Hii hupunguza kwa kiasi kikubwa matumizi ya ziada ya rasilimali, hurahisisha kutabiri tabia ya mfumo, huongeza kasi, na huwapa wasanidi uzoefu ulio rahisi zaidi na usio na utata. Kampuni inakataa waziwazi hali fiche za mfumo, mifumo inayoficha jinsi mambo yanavyofanya kazi, na mazingira ya uendeshaji yanayotegemea sana Python, ikipendelea uwazi, uwezo wa kuunganisha vipengele, na utendaji. ZML hivi karibuni ilitoa ZML/LLMD, seva yenye uwezo mkubwa wa inferensi ya LLM, na tayari imepata kutambuliwa katika jumuiya ya AI, ikiwemo kuungwa mkono hadharani na mshindi wa Turing Award, Yann LeCun. Mfumo huu umejengwa mahususi kwa mazingira ya uzalishaji ambako utendaji na uwezo wa kutumia maunzi mbalimbali ni muhimu zaidi. ZML inafaa sana kwa wahandisi wa AI, timu za ML, kampuni changa, na makampuni makubwa yanayotumia modeli kwa kiwango kikubwa na yanayotaka kuepuka kufungiwa kwa mtoa huduma mmoja huku yakifikia kasi ya juu zaidi inayowezekana ya inferensi kwenye chipu tofauti. Mambo Muhimu Inferensi isiyofungamana kabisa na aina maalumu ya maunzi kwa kutumia msingi mmoja wa msimbo. Ukusanyaji wa msimbo moja kwa moja kwa ajili ya vichapuzi mbalimbali ili kufikia utendaji wa juu zaidi. Matumizi madogo sana ya ziada ya rasilimali na uzoefu bora kwa wasanidi. Msisitizo mkubwa katika utayari wa kutumika katika uzalishaji na urahisi wa kutabiri tabia https://www.youtube.com/embed/Oguct9Y5QK8 2026-11-10T23:32:59+00:00 https://quasa.io/nd/video/prime-intellect-qeqetsha-i-ai-esezingeni-eliphezulu-kakhulu-ngaphandle-kwe-big-tech-ingcazelo-egcweleyo 2026-11-10T23:33:01+00:00 https://i.ytimg.com/vi/On47OM1RTZc/hqdefault.jpg Prime Intellect: Qeqetsha i-AI Esezingeni Eliphezulu Kakhulu Ngaphandle kwe-Big Tech (Ingcazelo Egcw #Quasa #QUA #PrimeIntellect I-Prime Intellect yakha i-Open Superintelligence Stack — inkundla ehlanganisa zonke izingxenye zobuchwepheshe evumela loba ngubani ukuthi aqeqetshe, afake ekusebenzeni njalo aqhubeke ethuthukisa amamodeli akhe e-AI ngaphandle kokuthembela kungqalasizinda ye-Big Tech. Inhloso yenkampani yikwenza i-AI esezingeni eliphezulu kakhulu itholakale lula njalo ibe ngeyabantu bebonke ngokusebenzisa amandla okubala ngamakhomphyutha angalawulwa yisikhungo esisodwa, amathuluzi alekhodi yomthombo evulelekileyo kanye lentuthuko ekhokhelwa ngumphakathi. Okuyisisekelo se-Prime Intellect yindawo yokusebenza ehlanganisiweyo ehlanganisa amandla okubala ngamakhomphyutha, ukuqeqetsha ngokufunda ngokuqiniswa (RL), ukuhlola, ukwenza iziphetho kanye lezindawo ezivikelekileyo zokuhlola ezehlukanisiweyo. Abasebenzisi bangafinyelela izinkulungwane zezindawo ze-RL ezakhiwe ngumphakathi nge-Environment Hub, baqhube ukuqeqetsha okukhulu okubanjelwe engqalasizinda yenkundla, bafake amamodeli ekusebenzeni ngokusekelwa kwe-LoRA okwakhelwe ngaphakathi, njalo basebenzise indawo yokuthengiselana yomhlaba wonke yama-GPU, kusukela kumanodi angawodwa kusiya kumaqoqo amakhulukazi. Inkampani isivele ifinyelele impumelelo emangalisayo kwezobuchwepheshe. Yakhupha i-INTELLECT-2, imodeli yokuqala elamapharamitha angu-32B eyaqeqetshwa ngokufunda ngokuqiniswa okusabalaliswe emhlabeni wonke, yalandelwa yi-INTELLECT-3 (imodeli ye-Mixture-of-Experts engu-100B+) le-SYNTHETIC-2 (izigidi zemilandela yezinyathelo zokucabanga ezakhiwe ngokubambisana). Amalabhulali abo alekhodi yomthombo evulelekileyo — i-Verifiers le-Prime-RL — enza ukuthi i-RL esabalalisiweyo yenzeke ngobukhulu, izinqubo zayo zingalindelani njalo iqhubeke isebenza lanxa kuleziphambeko. Isekelwa ngabakhokhelayo kwezokutshalwa kwezimali labalombono ngekusasa le-AI, abagoqela i-Founders Fund, i-Radical Ventures, i-NVIDIA, i-Intel, u-Andrej Karpathy, u-John Schulman labanye, i-Prime Intellect ihlanganisa ucwaningo oluphambili lengqalasizinda esebenzisekayo. Amakhasimende afana https://www.youtube.com/embed/On47OM1RTZc 2026-11-10T23:33:01+00:00 https://quasa.io/eu/video/zml-aztergai-errendimendu-gorena-edozein-txipetan-nvidia-amd-tpu-eta-gehiago 2026-11-10T23:33:05+00:00 https://i.ytimg.com/vi/Oguct9Y5QK8/hqdefault.jpg ZML aztergai: errendimendu gorena edozein txipetan (NVIDIA, AMD, TPU eta gehiago) #Quasa #QUA #zml ZML Parisen egoitza duen adimen artifizialeko azpiegituren enpresa bat da, eta adimen artifizialeko lan-kargak hardware jabedunetik bereizten dituen ekoizpenerako moduko inferentzia-pila garatzen ari da. Enpresaren oinarrizko filosofiak — “Ereduetik metalera” — aukera ematen du edozein eredu hainbat azeleragailutan (NVIDIA, AMD, TPU, Trainium eta gehiago) exekutatzeko kode-base bakar batetik abiatuta, hardwarearen errendimendu gorena lortuz. Python exekuzio-ingurune astunetan eta abstrakzio-geruzetan oinarritzen diren ohiko frameworkek ez bezala, ZMLk ereduak hardwarerako zuzenean konpilatzen ditu, Zig eta MLIR erabiliz. Horri esker, baliabide-gainkarga nabarmen txikiagoa da, aurreikusgarritasuna hobea, abiadura handiagoa eta garatzailearen esperientzia askoz garbiagoa. Enpresak berariaz baztertzen ditu ezkutuko egoera, abstrakzio magikoak eta Pythonen mendekotasun handia duten exekuzio-inguruneak, esplizitutasunaren, konposagarritasunaren eta errendimenduaren alde egiteko. ZMLk duela gutxi ZML/LLMD kaleratu du, LLM inferentziarako zerbitzari indartsu bat, eta dagoeneko aitortza lortu du adimen artifizialaren komunitatean; besteak beste, Yann LeCun Turing sariaren irabazleak babesa eman dio publikoki. Pila berariaz diseinatuta dago errendimenduak eta hardwarearen malgutasunak garrantzirik handiena duten ekoizpen-inguruneetarako. ZML aproposa da ereduak eskala handian hedatzen dituzten adimen artifizialeko ingeniarientzat, ikaskuntza automatikoko taldeentzat, enpresa sortu berrientzat eta enpresa handientzat, baldin eta hornitzaile bakarrari lotuta geratzea saihestu eta, aldi berean, txip desberdinetan ahalik eta inferentzia-abiadurarik handiena lortu nahi badute. Alderdi nabarmenenak Hardwarearekiko benetako independentzia duen inferentzia, kode-base bakarrarekin. Zuzeneko konpilazioa hainbat azeleragailutarako, errendimendu gorena lortzeko. Baliabide-gainkarga minimoa eta garatzailearentzako esperientzia bikaina. Arreta berezia ekoizpenean erabiltzeko prestasunari eta aurreikusgarritasunari. Kon https://www.youtube.com/embed/Oguct9Y5QK8 2026-11-10T23:33:05+00:00 https://quasa.io/lb/video/zml-am-test-maximal-leeschtung-op-all-chip-nvidia-amd-tpu-a-mei 2026-11-10T23:33:47+00:00 https://i.ytimg.com/vi/Oguct9Y5QK8/hqdefault.jpg ZML am Test: Maximal Leeschtung op all Chip (NVIDIA, AMD, TPU a méi) #Quasa #QUA #zml ZML ass eng Firma fir KI-Infrastruktur mat Sëtz zu Paräis, déi en Inferenzstack fir de produktiven Asaz entwéckelt, deen d'KI-Aarbechtslaaschte vu proprietärer Hardware entkoppelt. Hir zentral Philosophie — “Vum Modell bis bei d'Metall” — erméiglecht et, all Modell op verschiddenen Accelerateuren (NVIDIA, AMD, TPU, Trainium a méi) mat enger eenzeger Codebasis auszeféieren a gläichzäiteg déi maximal Leeschtung vun der Hardware ze liwweren. Am Géigesaz zu traditionelle Frameworks, déi op schwéier Python-Lafzäitëmfeld a verschidden Abstraktiounsschichte setzen, kompiléiert ZML Modeller mat Zig an MLIR direkt fir d'Hardware. Dat féiert zu däitlech manner zousätzlechem Ressourcëverbrauch, besserer Viraussehbarkeet, méi héijer Vitess an engem vill méi kloren Entwécklungserliefnes. D'Firma leent verstoppt Zoustänn, magesch Abstraktiounen a staark op Python baséierend Lafzäitëmfeld ausdrécklech of a setzt amplaz op explizitt Strukturen, Kombinéierbarkeet a Leeschtung. ZML huet viru Kuerzem ZML/LLMD verëffentlecht, e leeschtungsstaarken LLM-Inferenzserver, a krut schonn Unerkennung an der KI-Communautéit, dorënner ëffentlech Ënnerstëtzung vum Turing Award-Gewënner Yann LeCun. De Stack ass speziell fir Produktiounsëmfeld entwéckelt, an deene Leeschtung a Flexibilitéit bei der Hardware am meeschten zielen. ZML ass ideal fir KI-Ingenieuren, ML-Ekippen, Startuppen an Entreprisen, déi Modeller a groussem Moossstaf asetzen an eng Ofhängegkeet vun engem eenzegen Ubidder vermeide wëllen, wärend si déi héchstméiglech Inferenzvitess op verschiddene Chips erreechen. Héichpunkten Wierklech hardwareonofhängeg Inferenz mat enger eenzeger Codebasis. Direkt Kompiléierung fir verschidden Accelerateure fir maximal Leeschtung. Minimalen zousätzleche Ressourcëverbrauch an en exzellent Entwécklungserliefnes. Staarke Fokus op d'Bereetschaft fir de produktiven Asaz an op Viraussehbarkeet. Méiglech Aspekter, déi ze berücksichtege sinn Als relativ nei an technesch déifgräifend Léisung kann ZML fir Ekippen, déi u Python-zentréiert https://www.youtube.com/embed/Oguct9Y5QK8 2026-11-10T23:33:47+00:00 https://quasa.io/mt/video/zml-analizi-l-oghla-prestazzjoni-fuq-kwalunkwe-cippa-nvidia-amd-tpu-u-ohrajn 2026-11-10T23:33:49+00:00 https://i.ytimg.com/vi/Oguct9Y5QK8/hqdefault.jpg ZML: Analiżi — L-ogħla prestazzjoni fuq kwalunkwe ċippa (NVIDIA, AMD, TPU u oħrajn) #Quasa #QUA #zml ZML hija kumpanija tal-infrastruttura tal-AI bbażata f’Pariġi li qed tibni ġabra ta’ saffi tas-softwer għall-inferenza lesta għall-użu fil-produzzjoni, li tifred il-kompiti tal-AI mill-ħardwer proprjetarju. Il-filosofija ewlenija tagħha — “Mill-mudell sal-metall” — tippermetti li kwalunkwe mudell jitħaddem fuq diversi aċċeleraturi (NVIDIA, AMD, TPU, Trainium u oħrajn) minn bażi waħda ta’ kodiċi filwaqt li tikseb l-ogħla prestazzjoni tal-ħardwer. Għall-kuntrarju tal-oqfsa tradizzjonali li jiddependu fuq ambjenti ta’ eżekuzzjoni tqal ta’ Python u saffi ta’ astrazzjoni, ZML tikkompila l-mudelli direttament għall-ħardwer permezz ta’ Zig u MLIR. Dan iwassal għal piż addizzjonali tal-ipproċessar ferm iżgħar, prevedibbiltà aħjar, veloċità ogħla, u esperjenza ferm aktar ċara għall-iżviluppaturi. Il-kumpanija tiċħad b’mod espliċitu l-istat moħbi, l-astrazzjonijiet maġiċi, u l-ambjenti ta’ eżekuzzjoni li jiddependu ħafna fuq Python, u tippreferi ċ-ċarezza espliċita, il-ħila li jingħaqdu komponenti, u l-prestazzjoni. ZML dan l-aħħar ħarġet ZML/LLMD, server b’saħħtu għall-inferenza tal-LLM, u diġà kisbet rikonoxximent fil-komunità tal-AI, inkluż appoġġ pubbliku minn Yann LeCun, rebbieħ tat-Turing Award. Il-ġabra ta’ saffi tas-softwer hija mibnija apposta għal ambjenti ta’ produzzjoni fejn il-prestazzjoni u l-flessibbiltà tal-ħardwer huma l-aktar importanti. ZML hija ideali għall-inġiniera tal-AI, għat-timijiet tal-ML, għan-negozji ġodda, u għall-intrapriżi li jħaddmu mudelli fuq skala kbira u jridu jevitaw li jintrabtu ma’ fornitur wieħed filwaqt li jiksbu l-ogħla veloċità possibbli tal-inferenza fuq ċipep differenti. Punti ewlenin Inferenza tassew indipendenti mill-ħardwer b’bażi waħda ta’ kodiċi. Kompilazzjoni diretta għal diversi aċċeleraturi biex tinkiseb l-ogħla prestazzjoni. Piż addizzjonali minimu tal-ipproċessar u esperjenza eċċellenti għall-iżviluppaturi. Enfasi qawwija fuq it-tħejjija għall-użu fil-produzzjoni u l-prevedibbiltà. Aspetti li jista’ jkun hemm bżonn jitqiesu Bħala soluzzjoni relattivam https://www.youtube.com/embed/Oguct9Y5QK8 2026-11-10T23:33:49+00:00 https://quasa.io/ga/video/zml-leirmheas-buaicfheidhmiocht-ar-aon-slis-nvidia-amd-tpu-tuilleadh 2026-11-10T23:33:58+00:00 https://i.ytimg.com/vi/Oguct9Y5QK8/hqdefault.jpg ZML — Léirmheas: Buaicfheidhmíocht ar Aon Slis (NVIDIA, AMD, TPU & Tuilleadh) #Quasa #QUA #zml Is cuideachta bonneagair AI atá lonnaithe i bPáras é ZML. Tá cruach bogearraí tátail á forbairt aige atá oiriúnach do thimpeallachtaí táirgthe agus a scarann ualaí oibre AI ó chrua-earraí dílseánaigh. Leis an mbunfhealsúnacht atá aige — “Ón tSamhail go dtí an Miotal” — is féidir aon samhail a rith ar roinnt luasairí (NVIDIA, AMD, TPU, Trainium agus tuilleadh) ó bhonn cóid amháin agus buaicfheidhmíocht na gcrua-earraí a bhaint amach ag an am céanna. Murab ionann agus creataí traidisiúnta a bhíonn ag brath ar thimpeallachtaí troma rite Python agus ar shraitheanna astarraingthe, tiomsaíonn ZML samhlacha go díreach do na crua-earraí trí Zig agus MLIR a úsáid. Fágann sé sin go mbíonn an forchostas i bhfad níos ísle, an t-iompar níos intuartha, an luas níos airde agus eispéireas an fhorbróra i bhfad níos simplí agus níos soiléire. Diúltaíonn an chuideachta go sainráite do staid fholaithe, d’astarraingtí draíochta agus do thimpeallachtaí rite a bhíonn ag brath go mór ar Python, agus tugann sí tús áite don tsoiléireacht, don chumas comhpháirteanna a chur le chéile agus don fheidhmíocht. D’eisigh ZML ZML/LLMD le déanaí, freastalaí cumhachtach tátail LLM, agus tá aitheantas faighte aige cheana féin i bpobal AI, lena n-áirítear tacaíocht phoiblí ó Yann LeCun, buaiteoir Ghradam Turing. Tá an chruach deartha go sonrach do thimpeallachtaí táirgthe ina bhfuil an fheidhmíocht agus solúbthacht na gcrua-earraí thar a bheith tábhachtach. Tá ZML an-oiriúnach d’innealtóirí AI, d’fhoirne ML, do ghnólachtaí nuathionscanta agus d’fhiontair a chuireann samhlacha i bhfeidhm ar mhórscála agus ar mian leo spleáchas ar sholáthraí amháin a sheachaint agus an luas tátail is airde is féidir a bhaint amach ar shliseanna éagsúla. Príomhghnéithe Tátal atá fíor-neamhspleách ar na crua-earraí le bonn cóid amháin. Tiomsú díreach do roinnt luasairí chun buaicfheidhmíocht a bhaint amach. Forchostas íosta agus eispéireas den scoth don fhorbróir. Béim láidir ar ullmhacht don táirgeadh agus ar intuarthacht. Nithe a d’Fhéadfadh a Bheith le https://www.youtube.com/embed/Oguct9Y5QK8 2026-11-10T23:33:58+00:00 https://quasa.io/cy/video/zml-adolygiad-perfformiad-brig-ar-unrhyw-sglodyn-nvidia-amd-tpu-a-mwy 2026-11-10T23:34:03+00:00 https://i.ytimg.com/vi/Oguct9Y5QK8/hqdefault.jpg ZML — Adolygiad: Perfformiad brig ar unrhyw sglodyn (NVIDIA, AMD, TPU a mwy) #Quasa #QUA #zml Mae ZML yn gwmni seilwaith AI sydd wedi'i leoli ym Mharis ac sy'n adeiladu pentwr casgliadu o safon cynhyrchu sy'n datgysylltu llwythi gwaith AI oddi wrth galedwedd perchnogol. Mae eu hathroniaeth graidd — “O'r Model i'r Metel” — yn caniatáu i unrhyw fodel redeg ar sawl cyflymydd (NVIDIA, AMD, TPU, Trainium a mwy) o un sylfaen god, gan sicrhau perfformiad brig y caledwedd ar yr un pryd. Yn wahanol i fframweithiau traddodiadol sy'n dibynnu ar amgylcheddau rhedeg Python trwm a haenau o haniaethu, mae ZML yn crynhoi modelau'n uniongyrchol ar gyfer y caledwedd gan ddefnyddio Zig ac MLIR. Mae hyn yn arwain at orbenion sylweddol is, mwy o ragweladwyedd, cyflymder uwch, a phrofiad llawer mwy syml a threfnus i ddatblygwyr. Mae'r cwmni'n gwrthod yn benodol gyflwr cudd, haniaethiadau hudol, ac amgylcheddau rhedeg sy'n dibynnu'n helaeth ar Python, gan ffafrio eglurder, y gallu i gyfuno cydrannau, a pherfformiad. Yn ddiweddar, rhyddhaodd ZML ZML/LLMD, gweinydd casgliadu LLM pwerus, ac mae eisoes wedi ennill cydnabyddiaeth yn y gymuned AI, gan gynnwys cefnogaeth gyhoeddus gan Yann LeCun, enillydd Gwobr Turing. Mae'r pentwr wedi'i adeiladu'n benodol ar gyfer amgylcheddau cynhyrchu lle mae perfformiad a hyblygrwydd caledwedd o'r pwys mwyaf. Mae ZML yn ddelfrydol ar gyfer peirianwyr AI, timau ML, busnesau newydd, a mentrau sy'n rhoi modelau ar waith ar raddfa fawr ac sydd am osgoi cael eu clymu wrth un cyflenwr, gan sicrhau'r cyflymder casgliadu uchaf posibl ar draws gwahanol sglodion. Uchafbwyntiau Casgliadu sy'n wirioneddol annibynnol ar galedwedd gydag un sylfaen god. Crynhoyn uniongyrchol ar gyfer sawl cyflymydd i sicrhau perfformiad brig. Gorbenion lleiaf posibl a phrofiad rhagorol i ddatblygwyr. Pwyslais cryf ar barodrwydd ar gyfer cynhyrchu a rhagweladwyedd. Ystyriaethau posibl Gan ei fod yn ddatrysiad cymharol newydd sy'n dechnegol gymhleth, efallai y bydd angen cyfnod o ddysgu ar dimau sy'n gyfarwydd â fframweithiau sy'n canolbwyntio ar Python. Mae'n fwyaf addas i ddefnyddwyr sy'n rhoi blaenoriaeth i be https://www.youtube.com/embed/Oguct9Y5QK8 2026-11-10T23:34:03+00:00 https://quasa.io/fy/video/zml-resinsje-topprestaasjes-op-elke-chip-nvidia-amd-tpu-mear 2026-11-10T23:34:30+00:00 https://i.ytimg.com/vi/Oguct9Y5QK8/hqdefault.jpg ZML-resinsje: Topprestaasjes op elke chip (NVIDIA, AMD, TPU & mear) #Quasa #QUA #zml ZML is in AI-ynfrastruktuerbedriuw út Parys dat in foar produksje geskikte ynferinsjestack bout dy't AI-wurklêsten loskeppelt fan sletten hardware. Harren kearnfilosofy — “Fan model nei metaal” — makket it mooglik om elk model út ien koadebasis wei op meardere fersnellers (NVIDIA, AMD, TPU, Trainium en mear) draaie te litten en dêrby de maksimale prestaasjes út de hardware te heljen. Oars as tradisjonele raamwurken dy't op swiere Python-útfieromjouwings en abstraksjelagen fertrouwe, kompilearret ZML modellen streekrjocht nei de hardware mei Zig en MLIR. Dat soarget foar gâns minder ekstra belesting, bettere foarsisberens, hegere faasje en in folle oersichtliker ûntwikkelûnderfining. It bedriuw wiist ferburgen tastân, magyske abstraksjes en útfieromjouwings dy't swier op Python leune útdruklik ôf en kiest foar eksplisytens, gearstalberens en prestaasjes. ZML hat koartlyn ZML/LLMD útbrocht, in krêftige LLM-ynferinsjetsjinner, en hat al erkenning krigen yn de AI-mienskip, ûnder oare mei iepenbiere stipe fan Turing Award-winner Yann LeCun. De stack is spesjaal boud foar produksjeomjouwings dêr't prestaasjes en fleksibiliteit yn hardware it swierst weagje. ZML is ideaal foar AI-yngenieurs, ML-teams, startende bedriuwen en grutte ûndernimmingen dy't modellen op grutte skaal ynsette en ôfhinklikens fan ien leveransier foarkomme wolle, wylst se de heechst mooglike ynferinsjefaasje op ferskillende chips berikke. Hichtepunten Wier hardwareûnôfhinklike ynferinsje mei ien koadebasis. Streekrjochte kompilaasje nei meardere fersnellers foar topprestaasjes. Minimale ekstra belesting en in treflike ûntwikkelûnderfining. Sterke klam op reeëns foar produksje en foarsisberens. Mooglike omtinkenpunten As relatyf nije en technysk yngeande oplossing kin it foar teams dy't wend binne oan raamwurken mei Python as middelpunt wat leartiid fergje. It meast geskikt foar brûkers dy't suvere prestaasjes en kontrôle foaropstelle. Einoardiel: 4.5/5 stjerren ZML falt op as ien fan de meast ambisjeuze en technysk sekuere oanpakke https://www.youtube.com/embed/Oguct9Y5QK8 2026-11-10T23:34:30+00:00