Kwenzekeni
I-Ant Group enemithombo evulekile ye-Ling-3.0-flash-Fin, imodeli yepharamitha ye-Mixture-of-Experts (MoE) yebhiliyoni engu-124 eyenza kusebenze amapharamitha ayizigidi eziyizinkulungwane ezingu-5.1 ithokheni ngayinye. Imodeli yakhelwe ngokukhethekile ukugeleza komsebenzi wocwaningo lwezezimali, ukuhlanganisa ukubuyiswa kolwazi, ukumodela ukulinganisa nge-Excel automation, kanye nokukhiqizwa kwemibiko. Itholakala ku-OpenRouter, Vercel, Hugging Face, naku-ModelScope. Ngasikhathi sinye, i-Ant Group ikhiphe i-FinFIRST, ibhentshimakhi yabenzeli bokusesha ngezezimali eyakhiwe ne-China International Capital Corporation, ehlola inqubo yocwaningo kunezimpendulo zokugcina nje.
I-Ant Group ikhiphe i-Ling-3.0-flash-Fin, imodeli ye-AI enesisindo esivulekile eyenzelwe ucwaningo lwezezimali nokuhlaziya. Imodeli isebenzisa i-Architecture ye-Mixture-of-Experts (i-MoE) enepharamitha eyizigidi eziyizinkulungwane ezingu-124, yenza kusebenze izigidigidi ezi-5.1 kuphela ithokheni ngayinye ukuze kulinganiswe umthamo wolwazi nokusebenza kahle kokucabanga. Lo mklamo uhlose ukunciphisa izindleko zokuthunyelwa ngenkathi kugcinwa ukusebenza okuphezulu emisebenzini yezezimali.
Imodeli igxile emandleni amane ayinhloko: ukubuyiswa kolwazi oluvela emithonjeni enegunya nokulandeleka kokugcina, ukucabanga kocwaningo okwakha amaketango obufakazi obungaqinisekiswa kusukela kudatha ehlukahlukene, ukumodela ukulinganisa okuqondayo nokwenza ngokuzenzakalelayo amamodeli ezezimali e-Excel, kanye nokukhiqizwa kombiko okuhlanganisa amaqiniso, izibalo, namashadi kube imiphumela ehlelekile. Le ndlela egxile ekuhambeni komsebenzi iyayihlukanisa kumamodeli engxoxo enhloso evamile.
Ukutholakala kubanzi, ngemodeli efinyeleleka ngabahlinzeki be-API i-OpenRouter ne-Vercel, nezisindo ezivulekile ezitholakala ku-Hugging Face naku-ModelScope. Lokhu kuvumela abathuthukisi ukuthi bakhiphe imodeli ngasese futhi bayihlanganise nosesho lwangokwezifiso, i-Python, isizindalwazi, nokugeleza komsebenzi wesipredishithi. Ukukhishwa kuyingxenye yomndeni obanzi we-Ling 3.0, ohlanganisa amanye amamodeli wama-ejenti okukhiqiza kanye nemithwalo yemisebenzi eminingi.
Eceleni kwemodeli, i-Ant Group enemithombo evulekile ye-FinFIRST, ibhentshimakhi yabasebenzeli bokusesha ngezezimali. Ithuthukiswe ngokusekelwa kochwepheshe okuvela e-China International Capital Corporation, i-FinFIRST V1 ihlanganisa imisebenzi egunyazwe ngochwepheshe engu-123, imibandela ye-athomu engu-701, namaphoyinti erubrikhi angu-12,300. Ngokungafani namabhentshimakhi endabuko ahlola izimpendulo zokugcina kuphela, i-FinFIRST ihlola inqubo yocwaningo ebanzi, okuhlanganisa ukungaguquguquki kwedatha nokulandeleka, ihlinzeka ngesilinganiso esibanzi kakhulu sokwethembeka kwe-AI ezimeni zezimali.
Imininingwane yomthombo: opensourceforu.com ↗
Kungani kubalulekile
Lokhu kukhishwa kubhekana negebe elibalulekile ekusetshenzisweni kwe-AI kwezezimali: isidingo sokulandeleka, okuphumayo okubalayo okunembe kunokukhiqiza umbhalo ojwayelekile. Ngokuhlanganisa ukusebenza kahle kwe-MoE namathuluzi ezezimali akhethekile, yehlisa umgoqo wokusatshalaliswa okuyimfihlo kwamamodeli asezingeni eliphezulu ezindaweni ezilawulwayo. Ukufakwa kwebhentshimakhi egxile enqubweni (i-FinFIRST) ihlinzeka ngezinga elisha lokuhlola ukwethembeka kwe-AI ekuhlaziyweni kwezimali, kudlulele ngale kwamamethrikhi alula okunemba ukuze kuhlolwe ukuhambisana kwedatha namaketango obufakazi.
Umkhakha wezezimali udinga ukunemba okuphezulu kanye nokucwaningwa kwamabhuku, izindawo lapho ama-LLM enhloso ejwayelekile evamise ukudonsa kanzima. Ngokuhlanganisa i-spreadsheet automation kanye nokubuyiswa kolwazi olulandelekayo, i-Ling-3.0-flash-Fin iqondise amaphuzu athile obuhlungu ocwaningo lokutshalwa kwezimali, lapho amaphutha ekubaleni noma ekunikezeni umthombo angaba nemiphumela ebalulekile.
I-architecture ye-MoE ibalulekile ekwamukelweni kwebhizinisi ngoba ivumela ukuthunyelwa kwamamodeli amakhulu ane-overhead yekhompyutha ephansi. Lokhu kwenza kube lula ukuthi amafemu asebenzise lawa mamodeli emagcekeni noma ezindaweni eziyimfihlo zamafu, kubhekwana nokukhathazeka kobumfihlo bedatha okuvamile kwezezimali.
Ukukhishwa kwe-FinFIRST kuyintuthuko ephawulekayo ekuhlolweni kwe-AI. Ngokugxila enqubweni yocwaningo kunomphumela nje kuphela, inikeza uhlaka lokuhlola ukwethembeka kwama-ejenti e-AI emisebenzini yezimali eyinkimbinkimbi, enezinyathelo eziningi. Lokhu kungaba nomthelela ekutheni ezinye izinhlangano ziwahlola kanjani amathuluzi e-AI ngemisebenzi yebhizinisi ebalulekile.
Imvelo yesisindo esivulekile yemodeli kanye nebhentshimark ikhuthaza ukubeka izinto obala kanye negalelo lomphakathi. Ivumela abacwaningi abazimele kanye nabasebenzi ukuthi baqinisekise amakhono emodeli futhi babe namandla okuba ngcono phezu kwebhentshimakhi, ikhuthaze i-ecosystem eqine kakhudlwana ye-AI yezezimali.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
Which component of an AI application is the machine-learning model itself?
Ongakubuka ngokulandelayo
Gada ukutholwa kwamabhange otshalomali kanye nabaphathi bempahla ukuze kusetshenziswe ezimele. Buka ukuqinisekiswa okuzimele kokusebenza kwemodeli kubhentshimakhi ye-FinFIRST, njengoba imiphumela yamanje izibika yona Iqembu le-Ant. Qaphela uma ezinye izikhungo zezezimali zenza amamodeli afanayo akhethekile anesisindo esivulekile noma uma lokhu kuba indinganiso yokuhlolwa kwezimali kwe-AI.
Ukuqinisekiswa okuzimele kokusebenza kwemodeli ku-FinFIRST namanye amabhentshimakhi ezezimali ayashoda okwamanje, njengoba imiphumela ebikiwe iphuma emthonjeni ocaphuna izimangalo ze-Ant Group. Ukuhlola okuvela eceleni kuzobaluleka ukuze kuqinisekiswe ukusetshenziswa okusebenzayo kwemodeli.
Ukwamukelwa yizikhungo ezinkulu zezezimali kuzobonisa umthelela womhlaba wangempela wemodeli. Buka izehlakalo noma izimemezelo ezivela emabhange, abaphathi bempahla, noma izinkampani ze-fintech ezithumela i-Ling-3.0-flash-Fin ezindaweni zokukhiqiza.
Ukuvela kwebhentshimakhi ye-FinFIRST kufanelekile ukugadwa. Uma izuza amandla njengezinga lokuhlola i-AI yezezimali, ingalolonga ukuthuthukiswa kwamamodeli wesikhathi esizayo futhi isethe okulindelwe okusha ngokwethembeka kwe-AI kulo mkhakha.
Amazinga anamandla okulawula noma emboni mayelana nokusetshenziswa kwe-AI kwezezimali angase athonywe ukutholakala kwamamodeli anjalo akhethekile, alandelekayo. Idizayini yale modeli ihambisana nezidingo ezikhulayo zokucaciswa nokuhleleka ezinqumweni zezezimali eziqhutshwa yi-AI.