Kutshanje, i-LinSeer, indawo yabucala yeqonga elikhulu lokulinganisa elizimeleyo eliphuhliswe yi-H3C phantsi kolwalathiso lweQela le-Unisoc, lifumene i-4+ yokulinganisa kwimodeli yokuqinisekisa ukuthotyelwa kwemodeli yangaphambili ye-China Institute of Information Industry, ukufikelela ekhaya. inqanaba eliphezulu. eTshayina. Olu vavanyo lubanzi, olunamacala amaninzi lujolise kwiimodyuli ezisebenzayo zeLinSeer: ulawulo lwedatha, uqeqesho lwemodeli, ulawulo lwemodeli, ukuhanjiswa kwemodeli, kunye nenkqubo yophuhliso edibeneyo. Ibonisa amandla ahamba phambili e-H3C kwintsimi yemodeli enkulu kwicandelo labucala kwaye iya kunika inkxaso eqinile kumashishini ahlukeneyo ukuba angene kwixesha le-AIGC.
Njengoko ukuthandwa kwe-AIGC kuqhubeka nokunyuka, inkqubo yophuhliso lweemodeli ezinkulu ze-AI ziyakhawuleza, ngaloo ndlela kudala isidingo semigangatho. Ngokuphathelele kulo mba, i-Academy yase-China yoShishino loLwazi, ngokubambisana nezifundiswa, amaziko ophando lwenzululwazi, kunye noshishino, ikhuphe i-Trusted Artificial Intelligence Large-Scale Model Standard System 2.0. Le nkqubo yomgangatho ibonelela ngereferensi ebanzi yovandlakanyo lwezenzululwazi lwesakhono sobugcisa kunye nokusebenza ngokufanelekileyo kwemodeli yomgangatho omkhulu. I-H3C ithathe inxaxheba kolu vavanyo kwaye yavavanya ngokubanzi izakhono zophuhliso ze-LinSeer ukusuka kwizalathisi zovavanyo ezintlanu, ebonisa amandla ayo obugcisa abalaseleyo.
Ulawulo lwedatha: Uvavanyo lugxile ekuqhutyweni kwedatha kunye namandla olawulo lwenguqulo yeemodeli ezinkulu, kubandakanywa ukucocwa kwedatha, inkcazo, ukuhlolwa komgangatho, njl. Ngokusebenzisa ulawulo olululo lwedatha kunye nokulungiswa kwedatha, kudibaniswa nokufunyanwa komgangatho wedatha yeqonga le-Oasis, linokuxhasa ngokupheleleyo inkcazo yombhalo, umfanekiso, iaudio kunye nedatha yevidiyo.
Uqeqesho lomzekelo: Uvavanyo lujolise ekukwazini kweemodeli ezinkulu zokuxhasa iindlela ezininzi zoqeqesho, ukubonwa, kunye nokucwangciswa kokusetyenziswa kwezibonelelo. Ngokusekelwe kwiModeli njengeNkonzo (i-MaaS) yokwakhiwa kwezakhiwo, i-H3C ibonelela ngoqeqesho olubanzi lweemodeli ezinkulu kunye neenkonzo zokulungiswa kakuhle ukuvelisa imodeli elungiselelweyo kunye neyodwa kubathengi. Iziphumo zibonisa ukuba i-LinSeer ixhasa ngokupheleleyo uqeqesho lwe-multi-modal, imisebenzi yoqeqesho lwangaphambili, ulwimi lwendalo, kunye neelwimi zeprogram, kunye ne-avareji yokuchaneka kokuchaneka kwe-91.9% kunye nezinga lokusetyenziswa kwezixhobo ze-90%.
Ulawulo lwemodeli: Uvavanyo lujolise ekukwazini kweemodeli ezinkulu zokuxhasa ukugcinwa kwemodeli, ulawulo lwenguqulelo, kunye nolawulo lwelogi. Ukugcinwa kweVector yeLinSeer kunye nokufunyanwa kwenza iimodeli zikhumbule kwaye zixhase iimeko ezichanekileyo zokuphendula. Iziphumo zibonisa ukuba i-LinSeer inokuxhasa ngokupheleleyo imodeli yokugcina imodeli efana nolawulo lwenkqubo yefayile kunye nolawulo lomfanekiso, kunye nobuchule bolawulo lwenguqulelo efana nolawulo lwemetadata, ukugcinwa kobudlelwane, kunye nolawulo lwesakhiwo.
Ukusasazwa komzekelo: Vavanya amandla eemodeli ezinkulu zokuxhasa ukulungiswa kakuhle kwemodeli, ukuguqulwa, ukuthena kunye nobungakanani. I-LinSeer ixhasa ii-algorithms zohlengahlengiso ezahlukeneyo ukuze zihlangabezane needatha ezahlukeneyo kunye neemfuno zemodeli yabathengi beshishini. Ikwabonelela ngezakhono zokuguqula imodeli yeentlobo ezininzi. I-LinSeer ixhasa ukuthenwa kwemodeli kunye nobungakanani, ukufikelela kumanqanaba aphezulu ngokwemigaqo yokukhawulezisa i-latency kunye nokusetyenziswa kwememori.
Inkqubo yophuhliso edibeneyo: Uvavanyo lujolise kubuchule bophuhliso obuzimeleyo kwiimodeli ezinkulu. I-LinSeer idityaniswe nesixhobo sokubeka iliso se-ICT se-H3C esigcweleyo esigcweleyo ukudibanisa ngokufanelekileyo zonke izigaba zophuhliso lwemodeli ye-AI enkulu kunye nokubonelela ngeqonga lophuhliso olumanyeneyo kunye nezixhobo. Abathengi beshishini loncedo basebenze ngokufanelekileyo iimodeli ezinkulu kwindawo yabucala, bakha ngokukhawuleza usetyenziso olukrelekrele, kwaye bafezekise "inkululeko yokusebenzisa imodeli."
I-H3C isebenzisa i-AI kwisicwangciso SONKE kwaye idibanisa ubukrelekrele bokwenziwa kuluhlu olupheleleyo lwesoftware kunye neemveliso zehardware ukufezekisa itekhnoloji epheleleyo kunye nemeko epheleleyo. Ukongeza, i-H3C icebise i-AI yesicwangciso sokuxhotyiswa kweshishini LONKE, elijolise ekuqondeni ngokunzulu iimfuno zeshishini, ukudibanisa amandla e-AI kwizisombululo zokugqibela, kunye nokubonelela ngeenkonzo kumaqabane ukunceda ukuphuculwa okunengqondo kumashishini ahlukeneyo.
Ukuze kuqhutyekwe nokukhuthaza ukutsha kwesicelo sobukrelekrele bokwenziwa kunye nokuphunyezwa kwemizi-mveliso, i-H3C yasungula isisombululo se-AIGC ngokubanzi, igxile kwiqonga elivumelayo, iqonga ledatha, kunye neqonga lamandla ekhompyuter. Esi sisombululo sibanzi sihlangabezana ngokupheleleyo neemfuno zeemeko zoshishino zabasebenzisi kwaye sinceda abathengi ngokukhawuleza ukwakha iimodeli zesizinda sabucala kunye nokugxila kushishino, ukugxila kwingingqi, ukuba yedwa kwedatha, kunye nokuqhelaniswa nexabiso.
Ixesha lokuposa: Sep-22-2023
