Muva nje, i-LinSeer, isizinda esizimele sokumodela isizinda esikhulu esithuthukiswe ngokuzimele yi-H3C ngaphansi kokuqondisa kwe-Unisoc Group, ithole isilinganiso esingu-4+ ekuqinisekiseni ukuhambisana kwemodeli yangaphambi kokuqeqeshwa kwe-China Institute of Information Industry, okufinyelele ekhaya. izinga eliphezulu. China. Lokhu kuhlola okuphelele, okunezinhlangothi eziningi kugxile kumamojula amahlanu asebenzayo we-LinSeer: ukuphathwa kwedatha, ukuqeqeshwa kwamamodeli, ukuphathwa kwamamodeli, ukuthunyelwa kwamamodeli, kanye nenqubo yokuthuthukisa edidiyelwe. Ibonisa amandla ahamba phambili e-H3C emkhakheni wokumodela ngezinga elikhulu emkhakheni ozimele futhi izohlinzeka ngokusekela okuqinile kwezimboni ezihlukahlukene ukuze zingene enkathini ye-AIGC.
Njengoba ukuthandwa kwe-AIGC kuqhubeka nokukhuphuka, inqubo yokuthuthukiswa kwamamodeli amakhulu e-AI iyakhula, ngaleyo ndlela idale isidingo samazinga. Mayelana nalokhu, i-China Academy of Information Industry, ngokubambisana nezemfundo, izikhungo zocwaningo lwesayensi, kanye nezimboni, ikhiphe i-Trusted Artificial Intelligence Large-Scale Model Standard System 2.0. Lolu hlelo olujwayelekile luhlinzeka ngereferensi ebanzi yokuhlolwa kwesayensi kwamakhono obuchwepheshe nokusebenza kahle kokusebenza kwamamodeli amakhulu. I-H3C ibambe iqhaza kulokhu kuhlola futhi yahlola ngokujulile amakhono okuthuthukisa we-LinSeer kusukela kuzinkomba ezinhlanu zokuhlola, okubonisa amandla ayo obuchwepheshe amahle kakhulu.
Ukuphathwa kwedatha: Ukuhlola kugxile ekucutshungulweni kwedatha namandla okuphatha inguqulo amamodeli amakhulu, okuhlanganisa ukuhlanzwa kwedatha, isichasiselo, ukuhlolwa kwekhwalithi, njll. I-LinSeer ibonise ukusebenza okuhle kakhulu ekuhlanzeni idatha nokuphelela nokusekelwa komsebenzi. Ngokuphathwa kahle kwesethi yedatha nokucubungula idatha, kuhlanganiswe nokutholwa kwekhwalithi yedatha yesikhulumi se-Oasis, kungakwazi ukusekela ngokugcwele isichasiselo sombhalo, isithombe, umsindo, nedatha yevidiyo.
Ukuqeqeshwa kwamamodeli: Ukuhlola kugxile ekhonweni lamamodeli amakhulu okusekela izindlela eziningi zokuqeqesha, ukubona ngeso, kanye nokuhlelwa kokuthuthukisa izinsiza. Ngokusekelwe esakhiweni Semodeli Njengesevisi (i-MaaS), i-H3C inikeza ukuqeqeshwa okuphelele kwemodeli yezinga elikhulu kanye nezinsizakalo zokuhlela kahle ukuze kukhiqizwe amamodeli enziwe ngokwezifiso nakhethekile kumakhasimende. Imiphumela ibonisa ukuthi i-LinSeer isekela ngokugcwele ukuqeqeshwa kwezindlela eziningi, imisebenzi yangaphambi kokuqeqeshwa, ulimi lwemvelo, nezilimi zokuhlela, ngokunemba okukhulayo okumaphakathi okungu-91.9% kanye nezinga lokusebenzisa izinsiza lika-90%.
Ukuphathwa kwamamodeli: Ukuhlola kugxile ekhonweni lamamodeli amakhulu okusekela ukugcinwa kwamamodeli, ukuphathwa kwenguqulo, nokuphathwa kwamalogi. Ukugcinwa kwevekhtha ye-LinSeer nokubuyiswa kwenza amamodeli akhumbule futhi asekele izimo zokuphendula ezinembayo. Imiphumela ibonisa ukuthi i-LinSeer ingakwazi ukusekela ngokugcwele amakhono okulondoloza okuyimodeli njengokuphathwa kwesistimu yefayela nokuphathwa kwesithombe, kanye namandla okuphatha inguqulo njengokuphathwa kwemethadatha, ukugcinwa kobudlelwano, nokuphathwa kwesakhiwo.
Ukuthunyelwa kwemodeli: Linganisa ikhono lamamodeli amakhulu ukuze asekele ukuhleleka kahle kwemodeli, ukuguqulwa, ukuthena kanye nokulinganisa. I-LinSeer isekela ama-algorithms wokuhlela kahle ahlukahlukene ukuze ahlangabezane nezimo nedatha ehlukile nezidingo zemodeli zamakhasimende emboni. Iphinde inikeze ngamakhono okuguqula amamodeli ezinhlobo eziningi. I-LinSeer isekela ukuthenwa kwemodeli nokubalwa kwenani, ukufinyelela emazingeni athuthukile ngokuya ngokusheshiswa kwe-latency kanye nokusetshenziswa kwenkumbulo.
Inqubo yokuthuthukisa edidiyelwe: Ukuhlola kugxile emandleni okuthuthukisa azimele amamodeli amakhulu. I-LinSeer ihlanganiswe nethuluzi le-H3C lokuqapha ingqalasizinda ye-ICT egcwele isitaki ukuze ihlanganise zonke izigaba zokuthuthukiswa kwemodeli yezinga elikhulu le-AI futhi inikeze inkundla yokuthuthuka ehlangene namathuluzi. Siza amakhasimende embonini ukuthi avule ngempumelelo amamodeli amakhulu esizindeni esiyimfihlo, akhe ngokushesha izinhlelo zokusebenza ezihlakaniphile, futhi azuze “inkululeko yokusebenzisa imodeli.”
I-H3C isebenzisa i-AI kuwo wonke amasu futhi ihlanganisa ubuhlakani bokwenziwa ohlwini olugcwele lwesoftware nemikhiqizo yehadiwe ukuze kufinyelelwe ukumbozwa kobuchwepheshe bestaki esigcwele nobugcwele besimo. Ngaphezu kwalokho, i-H3C ihlongoze isu le-AI yazo YONKE imboni yokuhlomisa, okuhloswe ngayo ukuqonda ngokujulile izidingo zemboni, ukuhlanganisa amakhono e-AI ezixazululweni zokuphela-to-ekupheleni, nokuhlinzeka ngezinsizakalo kozakwethu ukuze basize ukuthuthukiswa okuhlakaniphile ezimbonini ezihlukahlukene.
Ukuze kuqhutshekwe nokukhuthaza ukusungulwa kwezicelo zobuhlakani bokwenziwa kanye nokuqaliswa kwezimboni, i-H3C yethule isisombululo sisonke se-AIGC, sigxile epulatifomu enika amandla, inkundla yedatha, kanye nenkundla yamandla ekhompuyutha. Lesi sixazululo esibanzi sihlangabezana ngokugcwele nezidingo zezimo zebhizinisi zabasebenzisi futhi sisiza amakhasimende akhe ngokushesha amamodeli esizinda esikhulu esizimele ngokugxila embonini, ukugxila kwesifunda, ukukhethekile kwedatha, nokuqondiswa kwenani.
Isikhathi sokuthumela: Sep-22-2023
