I-Nemotron 3: Ukubheja okukhulu kwe-NVIDIA okuvulekile kwe-AI yama-ejenti amaningi

Umncintiswano wobuhlakani bokwenziwa usuka kuma-chatbot alula, ahlukanisiwe uye ezinhlelweni zama-ejenti ezisebenzisana, eziphatha imisebenzi emide, futhi ezidinga ukuhlolwa. Kulesi simo esisha, I-NVIDIA inqume ukuthatha isinyathelo esicacile: ukuvula hhayi amamodeli kuphela, kodwa nedatha namathuluziukuze izinkampani, abaphathi bomphakathi kanye nezikhungo zocwaningo bakwazi ukwakha amapulatifomu abo e-AI ngokulawula okwengeziwe.

Lokho kunyakaza kwenzeka I-Nemotron 3, umndeni wamamodeli avulekile ahloselwe ukusebenzisa i-AI yama-ejenti amaningi Ifuna ukuhlanganisa ukusebenza okuphezulu, izindleko eziphansi zokucabanga, kanye nokucaca. Lesi siphakamiso asihloselwe nje njenge-chatbot nje evamile, kodwa njenge isisekelo sokuthumela amanxusa acabanga, ahlela futhi enze imisebenzi eyinkimbinkimbi emikhakheni elawulwayoLokhu kubaluleke kakhulu eYurophu naseSpain, lapho ubukhosi bedatha kanye nokuthobela imithetho kubalulekile khona.

Umndeni ovulekile wamamodeli we-AI e-ejenti kanye ne-ejensi

I-Nemotron 3 inikezwa njenge uhlelo oluphelele lwe-ecosystem: amamodeli, amasethi edatha, imitapo yolwazi, kanye nezindlela zokupheka zokuqeqesha ngaphansi kwamalayisense avulekile. Umqondo we-NVIDIA uwukuthi izinhlangano azisebenzisi nje i-AI njengesevisi engacacile, kodwa zingahlola okungaphakathi, zivumelanise amamodeli nezizinda zazo, futhi ziwasebenzise engqalasizinda yazo, kungaba sefwini noma ezikhungweni zedatha zasendaweni.

Le nkampani ihlela leli qhinga ngaphakathi kokuzibophezela kwayo kokuthi I-AI EbusayoOhulumeni nezinkampani eYurophu, eNingizimu Korea, nakwezinye izifunda bafuna ezinye izindlela ezivulekile esikhundleni sezinhlelo ezivaliwe noma zakwamanye amazwe, ezivame ukungavumelani kahle nemithetho yabo yokuvikela idatha noma izidingo zokuhlola. I-Nemotron 3 ihlose ukuba yisisekelo sobuchwepheshe sokwakha phezu kwaso amamodeli kazwelonke, emikhakha, noma ezinkampani ngokubonakala okukhulu nokulawula.

Ngokuhambisanayo, I-NVIDIA iqinisa isikhundla sayo ngale kwehadiweKuze kube manje, yayiwumhlinzeki we-GPU obhekisela kakhulu; nge-Nemotron 3, iphinde izibeke ohlakeni lwamathuluzi okumodela nokuqeqesha, incintisana ngqo nabadlali abanjengo-OpenAI, i-Google, i-Anthropic, noma i-Meta, futhi iphikisana namamodeli e-premium afana ne- I-SuperGrok EnzimaI-Meta ibilokhu inciphisa ukuzibophezela kwayo emthonjeni ovulekile ezizukulwaneni zakamuva ze-Llama.

Kuhlelo locwaningo lwaseYurophu kanye nohlelo lokuqalisa—oluthembele kakhulu kumamodeli avulekile abanjwe ezinkundleni ezifana ne-Hugging Face—ukutholakala kwezisindo, idatha yokwenziwa, kanye nemitapo yolwazi ngaphansi kwamalayisense avulekile kumelela enye indlela enamandla esikhundleni salokho Amamodeli aseShayina kanye namaMelika abusa ukuthandwa kanye nezinga lokulinganisa.

Ukwakhiwa kwe-MoE ye-Hybrid: ukusebenza kahle kwama-ejenti amakhulu

Isici sobuchwepheshe esiyinhloko seNemotron 3 yi- Ukwakhiwa kwe-hybrid kwengxubevange yochwepheshe efihliwe (MoE)Esikhundleni sokusebenzisa wonke amapharamitha emodeli ku-inference ngayinye, kuvuliwe ingxenye encane yawo kuphela, iqembu lochwepheshe elifanele kakhulu umsebenzi noma ithokheni okukhulunywa ngayo.

Le ndlela ivumela kunciphisa kakhulu izindleko zokubala kanye nokusetshenziswa kwememoriLokhu kwandisa nokusetshenziswa kwamathokheni. Kuma-architecture e-multi-agent, lapho amashumi noma amakhulu ama-agent eshintshana khona imiyalezo njalo, lokhu kusebenza kahle kuyisihluthulelo sokuvimbela uhlelo ukuthi lungaqhubeki ngokwezindleko ze-GPU kanye nefu.

Ngokusho kwedatha eyabiwe yi-NVIDIA kanye nezilinganiso ezizimele, i-Nemotron 3 Nano ifinyelela amathokheni afinyelela kane ngomzuzwana Uma kuqhathaniswa neyandulelayo, i-Nemotron 2 Nano, inciphisa ukukhiqizwa kwamathokheni okucabanga okungadingekile cishe ngo-60%. Empeleni, lokhu kusho izimpendulo ezilinganayo noma ezinembile kakhulu, kodwa "ngokusebenzisa amagama amaningi" kanye nezindleko eziphansi ngombuzo ngamunye.

Ukwakhiwa kwe-MoE okuhlanganisiwe, kuhlanganiswe namasu athile okuqeqesha, kuholele ekutheni Amamodeli amaningi avulekile athuthukile asebenzisa izinhlelo zochwephesheI-Nemotron 3 ijoyina lo mkhuba, kodwa igxile ngqo ku-AI e-ejenti: imizila yangaphakathi eyenzelwe ukuxhumanisa phakathi kwama-ejenti, ukusetshenziswa kwamathuluzi, ukuphathwa kwezimo ezinde, kanye nokuhlela isinyathelo ngesinyathelo.

Osayizi abathathu: i-Nano, i-Super, ne-Ultra yemisebenzi ehlukene

Umndeni wakwaNemotron 3 uhlelwe ngendlela osayizi abathathu abayinhloko bemodeli ye-MoE, zonke zivulekile futhi zinemingcele esebenzayo encishisiwe ngenxa yokwakhiwa kochwepheshe:

• I-Nemotron 3 Nano: cishe amapharamitha ayizigidigidi ezingama-30.000, cishe Izimpahla eziyizigidigidi ezingama-3.000 ngethokheni ngayinyeYenzelwe imisebenzi eqondiwe lapho ukusebenza kahle kubalulekile khona: ukulungisa amaphutha esofthiwe, ukufingqa amadokhumenti, ukubuyisa ulwazi, ukuqapha uhlelo, noma abasizi be-AI abakhethekile.
• I-Nemotron 3 Super: cishe amapharamitha ayizigidigidi eziyi-100.000, ane Izimpahla eziyizigidigidi eziyi-10.000 kuzo zonke izinyathelo. Ihloselwe Ukucabanga okuthuthukisiwe kuzakhiwo zama-ejenti amaningingokubambezeleka okuphansi ngisho nalapho ama-ejenti amaningi ebambisana ukuxazulula ukugeleza okuyinkimbinkimbi.
• I-Nemotron 3 Ultra: izinga eliphezulu, elinamapharamitha angaba yizigidigidi ezingu-500.000 futhi afinyelela ku- Izimpahla eziyizigidigidi ezingama-50.000 ngethokheni ngayinyeIsebenza njengenjini enamandla yokucabanga yocwaningo, ukuhlela amasu, ukwesekwa kwezinqumo ezisezingeni eliphezulu, kanye nezinhlelo ze-AI ezidinga kakhulu.

Empeleni, lokhu kuvumela izinhlangano Khetha usayizi wemodeli ngokuya ngesabelomali sakho kanye nezidingo zakhoI-Nano yemithwalo yemisebenzi emikhulu, enzima kanye nezindleko eziqinile; Kuhle kakhulu lapho kudingeka ukucabanga okujulile kakhulu nama-ejenti amaningi abambisanayo; kanye ne-Ultra yezehlakalo lapho ikhwalithi kanye nomongo omude kudlula khona izindleko ze-GPU.

Okwamanje I-Nemotron 3 Nano kuphela etholakalayo ukuze isetshenziswe ngokushesha.Izinhlobo ze-Super kanye ne-Ultra zihlelelwe ingxenye yokuqala ka-2026, okunikeza izinkampani zaseYurophu kanye nama-laboratory isikhathi sokuzama kuqala nge-Nano, ukusungula amapayipi, futhi kamuva, ukuhambisa amacala adinga amandla amakhulu.

I-Nemotron 3 Nano: iwindi lamathokheni ayisigidi esingu-1 kanye nezindleko eziqukethwe

I-Nemotron 3 Nano, njenganamuhla, iyasebenza. umholi osebenzayo womndeniI-NVIDIA iyichaza njengemodeli engabizi kakhulu ngokwezibalo ebangeni, elungiselelwe ukuletha ukusebenza okuphezulu emisebenzini yama-ejenti amaningi kanye nemisebenzi enzima kodwa ephindaphindwayo.

Phakathi kwezici zayo zobuchwepheshe, okulandelayo kuyagqama: ifasitela lomongo lamathokheni afinyelela esigidini esisodwaLokhu kuvumela ukugcinwa kwenkumbulo yamadokhumenti amaningi, izindawo zokugcina amakhodi eziphelele, noma izinqubo zebhizinisi ezinezinyathelo eziningi. Ezinhlelweni zaseYurophu ebhange, kwezempilo, noma ekuphathweni komphakathi, lapho amarekhodi angaba maningi khona, leli khono lokuqukethwe kwesikhathi eside libaluleke kakhulu.

Izilinganiso zenhlangano ezimele Ukuhlaziywa kokwenziwa kubeka i-Nemotron 3 Nano njengenye yamamodeli omthombo ovulekile alinganiselayo kakhulu Ihlanganisa ubuhlakani, ukunemba, kanye nesivinini, kanye namazinga okusebenzisa amathokheni amakhulu ngomzuzwana. Lokhu kuhlanganiswa kwenza kube kuhle kubahlanganisi be-AI kanye nabahlinzeki bezinsizakalo eSpain abadinga ulwazi oluhle lomsebenzisi ngaphandle kwezindleko ezinkulu zengqalasizinda.

Ngokuphathelene nezimo zokusetshenziswa, i-NVIDIA ihlose i-Nano ku- Isifinyezo sokuqukethwe, ukulungisa amaphutha esofthiwe, ukubuyisa ulwazi, kanye nabasizi be-AI yebhizinisiNgenxa yokwehliswa kwamathokheni okucabanga aphindaphindekile, kungenzeka ukusebenzisa ama-ejenti agcina izingxoxo ezinde nabasebenzisi noma izinhlelo ngaphandle kokunyuka okukhulu kwezindleko zokuqagela.

Idatha evulekile kanye nemitapo yolwazi: I-NeMo Gym, i-NeMo RL kanye noMhloli

Esinye sezici ezivelele kakhulu ze-Nemotron 3 ukuthi Akugcini ekukhululeni izisindo zamamodeliI-NVIDIA ihambisana nomndeni nesethi ephelele yezinsiza ezivulekile zokuqeqesha, ukulungisa, kanye nokuhlola ama-ejenti.

Ngakolunye uhlangothi, kwenza kutholakale i-corpus yokwenziwa amathokheni amaningana ayizigidigidi zedatha yangaphambi kokuqeqeshwa, yangemva kokuqeqeshwa, kanye nokuqinisaLawa masethi edatha, agxile ekucabangeni, ekubhaleni amakhodi, kanye nasekusebenzeni kwezinyathelo eziningi, avumela izinkampani nezikhungo zocwaningo ukuthi zikhiqize izinhlobo zazo zesizinda se-Nemotron (isb., ezomthetho, ezempilo, noma zezimboni) ngaphandle kokuqala phansi.

Phakathi kwalezi zinsizakusebenza, okulandelayo kuvelele: Isethi yedatha ye-Nemotron Agentic SafetyIqoqa idatha ye-telemetry ngokuziphatha kwe-ejenti ezimweni zangempela. Umgomo wayo ukusiza amaqembu ukulinganisa nokuqinisa ukuphepha kwezinhlelo ezizimele eziyinkimbinkimbi: kusukela kulokho okwenziwa yi-ejenti lapho ihlangana nedatha ebucayi, kuya endleleni esabela ngayo emiyalweni engacacile noma engaba yingozi.

Ngokuphathelene nesigaba samathuluzi, i-NVIDIA iyaqala I-NeMo Gym kanye ne-NeMo RL njengemitapo yolwazi evulekile ukuze kuqeqeshwe ngokuqiniswa kanye nokuqeqeshwa kwangemva kokuqeqeshwa, kanye noMhloli we-NeMo wokuhlola ukuphepha nokusebenza. Lawa matayibrari ahlinzeka ngezindawo zokulingisa ezilungele ukusetshenziswa kanye nezindlela zokulinganisa nomndeni wakwaNemotron, kodwa anganwetshwa nakwamanye amamodeli.

Konke lokhu—izisindo, amasethi edatha, kanye nekhodi—kusatshalaliswa ngokusebenzisa I-GitHub kanye ne-Hugging Face zinelayisensi ngaphansi kwe-NVIDIA Open Model License.ukuze amaqembu aseYurophu akwazi ukukuhlanganisa kalula kuma-MLOp awo. Izinkampani ezifana ne-Prime Intellect kanye ne-Unsloth sezivele zifaka i-NeMo Gym ngqo emisebenzini yazo yokusebenza ukuze kube lula ukufunda okuqinisiwe ku-Nemotron.

Ukutholakala emafwini omphakathi kanye nesistimu ye-ecosystem yaseYurophu

I-Nemotron 3 Nano isiyatholakala ku- Ubuso Obumbambayo y GitHubkanye nabahlinzeki be-inference abafana ne-Baseten, i-DeepInfra, i-Fireworks, i-FriendliAI, i-OpenRouter, kanye ne-Together AI. Lokhu kuvula ithuba lokuthi amaqembu okuthuthukisa eSpain ahlole imodeli nge-API noma ayithumele ezingqalasizinda zawo ngaphandle kobunzima obukhulu.

Ebusweni bamafu, I-Nemotron 3 Nano ijoyina i-AWS nge-Amazon Bedrock ngenxa yokucabanga okungenaseva, futhi imemezele ukusekela i-Google Cloud, i-CoreWeave, i-Crusoe, i-Microsoft Foundry, i-Nebius, i-Nscale, kanye ne-Yotta. Ezinhlanganweni zaseYurophu esezisebenza kulezi zinkundla, lokhu kwenza kube lula ukwamukela i-Nemotron ngaphandle kwezinguquko ezinkulu ekwakhiweni kwazo.

Ngaphezu kwefu lomphakathi, i-NVIDIA ikhuthaza ukusetshenziswa kwe-Nemotron 3 Nano njenge I-NIM microservice ingasetshenziswa kunoma iyiphi ingqalasizinda esheshisiwe ye-NVIDIALokhu kuvumela izimo ezixubile: ingxenye yomthwalo emafwini omhlaba wonke kanye nengxenye ezikhungweni zedatha zendawo noma emafwini aseYurophu abeka phambili ukuhlala kwedatha e-EU.

Izinguqulo I-Nemotron 3 Super ne-Ultra, eziqondiswe emisebenzini yokucabanga eyeqisayo kanye nezinhlelo ezinkulu zama-ejenti amaningi, kuhlelwe ingxenye yokuqala ka-2026Lo mugqa wesikhathi uvumela ucwaningo lwaseYurophu kanye nesistimu yebhizinisi isikhathi sokuzama i-Nano, ukuqinisekisa izimo zokusetshenziswa, kanye nokuklama amasu okufuduka aye kumamodeli amakhulu uma kudingeka.

I-Nemotron 3 ibeka i-NVIDIA njengomunye wabahlinzeki abaphambili be- amamodeli avulekile aphezulu aqondiswe ku-AI ejentiNgesiphakamiso esihlanganisa ukusebenza kahle kobuchwepheshe (i-hybrid MoE, i-NVFP4, umongo omkhulu), ukuvuleleka (izisindo, amasethi edatha kanye nemitapo yolwazi etholakalayo) kanye nokugxila okucacile ekubuseni kwedatha kanye nokucaca, izici ezibucayi kakhulu eSpain nakwamanye amazwe aseYurophu, lapho imithethonqubo kanye nengcindezi yokuhlola i-AI ikhula khona.

I-athikili ehlobene:

I-Microsoft Discovery AI iqhuba impumelelo yesayensi nezemfundo ngobuhlakani bokwenziwa komuntu siqu

U-Alberto Navarro

Ngingumshisekeli wezobuchwepheshe oguqule izintshisekelo zakhe "ze-geek" zaba umsebenzi. Ngichithe iminyaka engaphezu kwengu-10 yempilo yami ngisebenzisa ubuchwepheshe obusezingeni eliphezulu kanye nokukitaza ngazo zonke izinhlobo zezinhlelo ngenxa yelukuluku lokufuna ukwazi. Manje sengiqeqeshelwe ubuchwepheshe be-computer nemidlalo yama-video. Lokhu kungenxa yokuthi sekuphele iminyaka engaphezu kwengu-5 ngisebenza ngokubhalela amawebhusayithi ahlukahlukene ezobuchwepheshe nemidlalo yevidiyo, ngenza izindatshana ezifuna ukukunikeza imininingwane oyidingayo ngolimi oluqondakala yiwo wonke umuntu.

Uma unemibuzo, ulwazi lwami lusukela kuyo yonke into ehlobene nesistimu yokusebenza ye-Windows kanye ne-Android yomakhalekhukhwini. Futhi ukuzibophezela kwami ​​​​kuwe, ngihlala ngizimisele ukuchitha amaminithi ambalwa futhi ngikusize uxazulule noma yimiphi imibuzo ongase ube nayo kulo mhlaba we-inthanethi.