Kwenzekeni
I-QUASA ibika ukuthi ubufakazi bomqondo buka-August 2026 busebenzise isethi engaphansi egodliwe ye-MLCommons AILuminate ukwaziswa kokuphepha, ukubala okuvikelekile kwe-OpenMined kanye nemodeli efakwe esitsheni ye-Google DeepMind. Ukuhlelwa kuvimbele unjiniyela wemodeli ukuthi abone imiyalo yokuhlola futhi kwavimbela umhlinzeki webhentshimakhi kanye nomcwaningi mabhuku ukuthi abone izisindo zemodeli yobunikazi. I-QUASA ithi idizayini inciphisa ubungozi obuthile bokungcoliswa kanye nezindawo zobuhlakani, kodwa ayikuqedi ukuchayeka kwangaphambi kwesikhathi ezintweni zokuqhathanisa noma ukufakazela ukuthi ukuhlola kunokufaneleka kokwakha.
I-QUASA ibika ukuthi ukungcoliswa kwebhentshimakhi kungavela lapho imibuzo yokuhlola, izimpendulo noma okuhlukile kungena ukuqeqeshwa kwangaphambili, idatha yokulungisa kahle noma ukulungiselelwa okuphindaphindiwe. Isitolo sithi ukuphindaphinda okuqondile akudingekile: izimpendulo eziyingxenye, imiqondo ehlukile nezibonelo ezihlobene zingenza imisebenzi ejwayelekile ibe lula.
Ngokusho kwe-QUASA, umshayeli wendiza we-DeepMind–MLCommons usebenzise i-AVERI ukuze aqhube ukwaziswa kokuphepha kwe-MLCommons AILuminate egodliwe ngekhompyutha evikelekile ye-OpenMined kanye nemodeli ye-Google DeepMind Gemini Flash Lite efakwe esitsheni. Umnikazi wemodeli akakwazanga ukuhlola imibuzo egodliwe, kuyilapho i-MLCommons ne-AVERI ayikwazanga ukuhlola izisindo zobunikazi. Le mininingwane ibikwa yi-QUASA kusukela ku-akhawunti yokuqaliswa ecashuniwe futhi ayiqinisekiswanga ngokuzimela lapha.
I-outlet ihlukanisa ubuqotho bokwenza kusukela ekuqinisekiseni ukwakha. Esokuqala sithinta ukuthi izinto ezihlosiwe ezingabonwa kanye nesistimu emenyezelwe zisetshenziswe ngaphansi kwezimo ezilawulwayo. Okwesibili kumayelana nokuthi imisebenzi namamethrikhi asekela amandla, ukuphepha noma isimangalo sokwethembeka esinamathiselwe kumphumela.
I-QUASA ihlonza ukuvuza okusheshayo, ukuchayeka kwemodeli yesisindo, ukufinyelela komhloli kanye nokungcoliswa kokuqeqeshwa okuzayo njengezingozi ezihlukene. Ithi izilawuli ezinjengamarekhodi okufinyelela, imikhawulo yokugcinwa, imikhawulo yokuphumayo, izinqubo zesigameko kanye nemithetho yokuhoxiswa kwezinto eziveziwe kufanele ihambisane nokuvikelwa kwe-cryptographic.
Imininingwane yomthombo: quasa.io ↗
Kungani kubalulekile
Amaphuzu ebhentshimakhi e-AI aya ngokuya abe nomthelela ekuthengweni kwezimpahla, izimangalo zokuphepha nokuqhathaniswa phakathi kwamasistimu. Umbiko we-QUASA ugqamisa ukuthi ukugcinwa kuyimfihlo kubhekana nengxenye eyodwa kuphela yobuqotho bokuhlola: ukugcina imiyalo evikelekile kanye nezisindo eziyimodeli ngokuhlukana phakathi nesikhathi sokugijima. Ibhentshimakhi ingahlala ingameleli, ingatholi amaphuzu kahle noma isengozini yokungcoliswa esikhathini esizayo ngisho nalapho kungekho uhlangothi olubona okuyimfihlo kolunye. Lokho kwenza ukuvela kwento engabonakali, ukudalulwa kokucushwa, izilinganiso zokungaqiniseki nokuhlaziya ukwehluleka kubaluleke kakhulu kunoma ubani othembele emiphumeleni yokuhlolwa ye-AI.
Ukuhlola okuyimfihlo kungenza kube nzima ukuthi umthuthukisi oyimodeli alungiselele ngokuqondile ngokumelene nemibuzo egodliwe, kuyilapho futhi ekhawulela ukudalulwa kwezisindo zobunikazi. Lobo ubufakazi obusebenzisekayo mayelana nezimo zokuhlolwa, kodwa buncane kunobufakazi bokuthi imodeli iphephile, ithembekile noma iyakwazi ukuvama.
I-QUASA icaphuna ukubuyekezwa kwe-NeurIPS kwamabhentshimakhi angu-445 LLM njengokuthola ubuthakathaka obuqhubekayo kuzinto, imisebenzi kanye namamethrikhi wokuthola amaphuzu asetshenziselwa ukusekela izimangalo zebhentshimakhi. Umthombo awubanikezeli ababhali bokubuyekeza, indlela yokusebenza egcwele noma ukuhlola okuzimele kwemiphumela yomshayeli, ngakho leyo mininingwane ihlala ingaqinisekisiwe kulokhu kuhlola.
Okushiwo okungokoqobo ukuthi izinhlangano kufanele ziphathe amaphuzu angaboni kabili njengobufakazi obunemibandela. Badinga inguqulo yemodeli, ukulungiselelwa, ukwaziswa kwesistimu, amathuluzi, izilungiselelo zesampula, indawo yesikhathi sokusebenza, usayizi wesampula, izilinganiso zokungaqiniseki kanye nokwehluleka kwezinga lesigaba ngaphambi kokusebenzisa umphumela esinqumweni sokuphakelwa noma sokuthenga.
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
Buka ukuthi ingabe ukuhlola kwesikhathi esizayo kuyaqinisekisa yini ukuthi izinto ebezigodliwe bezivikelwe kanjani, ukuthi ukuchayeka kwaphenywa kanjani nokuthi kwenzekani ekwazisweni, kokuphumayo, kumalogi nedatha etholiwe ngemva kokuhlolwa. Amaqembu okuthenga kufanele futhi abuze ukuthi ingabe ukulungiselelwa kwemodeli ehloliwe kufana naleyo esetshenziswayo nokuthi ingabe ibhentshimakhi ibonisa ubungozi obuthile bokusetshenziswa okuhlosiwe. I-QUASA ayiqinisekisi ngokuhlola okuzimele ukuthi umshayeli wendiza okubikiwe uthuthukise ukufaneleka komhlaba wangempela noma ukuthi izilawuli zayo zivimbela yonke indlela yokuvuza.
Imibiko yesikhathi esizayo kufanele icacise ukuthi ubani ongafinyelela ukwaziswa, izisindo, izimpendulo, izimiso zamaphuzu, amalogi nobufakazi bokufakazela ngaphambi, phakathi nangemuva kokugijima.
Abanakekeli bebhentshimakhi kufanele bachaze ukuthi izinto zigodlwa kanjani, ukuthi ukuchayeka kuphenywa kanjani, ukuthi imibuzo esengozini ishintshwa kanjani nokuthi izinguqulo zebhentshimakhi zivuselelwa kanjani.
Izinhlangano kufanele ziqhathanise ukucushwa okuhloliwe nesistimu esetshenzisiwe futhi zihlole ukuthi imisebenzi imele isizinda sazo, okuhlanganisa ukuhluleka okungajwayelekile kodwa okulandelanayo.
I-QUASA ayibiki inani, ukutholakala komphakathi, inqubo yokufinyelela noma ukuthuthukiswa kokusebenza okulinganiselwe kokuhlola. Futhi ayiqinisekisi ngokuzimela ukuthi izilawuli ezichaziwe zivimbela konke ukuvuza noma ukungcoliswa kokuqeqeshwa komfula.