Qwen: Eyona modeli inomthombo ovulekileyo onokuyisebenzisa ngokwenene ngomlinganiselo omkhulu

#VideoAI #Web3 #Crypto

I-Qwen (eboniswe ku-Quasa.io/projects/qwen) lusapho lweemodeli ze-AI ezinomthombo ovulekileyo eziphambili kwezobuchwepheshe, ezivela kwi-Tongyi Lab yakwa-Alibaba, oluye lwaba ngomnye wabakhuphisana abanamandla kakhulu kugqatso lwe-AI lwehlabathi.

Iphuhliswe phantsi koluhlu lwe-Qwen (oluquka i-Qwen3.5, i-Qwen2.5-Max, i-Qwen-VL, neemodeli ezikhethekileyo ezifana ne-Qwen3-Coder), kwaye inika iziphumo ezibalaseleyo ekuqiqeni, ekubhaleni ikhowudi, ekuqondeni iintlobo ezahlukeneyo zolwazi, nakwizakhono zee-arhente, ngoxa iiparamitha zayo ezifunyenwe ngoqeqesho zihlala zivuleleke ngokupheleleyo kwaye isebenza ngeendleko eziphantsi kakhulu.

Embindini wayo, i-Qwen idibanisa ubukhulu obukhulu noyilo olukrelekrele (i-Mixture-of-Experts kwiinguqulelo ezinkulu) kunye noqeqesho olongezelelweyo olubanzi olusebenzisa ukufunda ngokuqiniswa.

Iimpawu eziphambili ziquka:

• I-Qwen3.5 — imodeli enamandla eneeparamitha ezivulelekileyo, enobuchule obubalaseleyo bokuqiqa, bokusingatha umxholo omde (ukuya kutsho kwii-token eziyi-1M+ kwezinye iinguqulelo), kunye nokusebenza ngamandla kwiilwimi ezininzi, ingakumbi kwisiTshayina nakwisiNgesi.
• I-Qwen-VL / Qwen-Image — izakhono eziphambili ezidibanisa ukubona nolwimi, ezinokuqonda, ukuvelisa nokuhlela imifanekiso ngokuchaneka okuphezulu.
• I-Qwen3-Coder — yenye yezona modeli zivulelekileyo zokubhala ikhowudi zibalaseleyo, egqwesa ekubhaleni ikhowudi ngee-arhente, ekusebenziseni izixhobo, nakwimisebenzi entsonkothileyo yobunjineli besoftware.
• Izakhono zee-arhente ezakhelwe ngaphakathi, ukusetyenziswa kwezixhobo okulungelelanayo, ii-arhente zophando olunzulu, kunye nezakhono zee-arhente ezisebenzisa ukubona ukwenza imisebenzi enamanyathelo amaninzi kwihlabathi lokwenene.
• Ukusebenza ngeendleko eziphantsi okukhuphisana ngamandla kakhulu (iindleko zokusebenzisa imodeli ukuvelisa iziphumo zidla ngokuba ngaphantsi nge-60% kunezo zezizukulwana zangaphambili) kunye neziphumo ezinamandla kwiimvavanyo zokuthelekisa, ezikhuphisana ne-GPT-4o ne-Claude 3.5 Sonnet okanye ezizodlulayo kwiindidi eziliqela.

Ibafanele ngokugqibeleleyo abaphuhlisi, abaphandi be-AI, amashishini asaqalayo, amashishini amakhulu, nabathandi bomthombo ovulekileyo abafuna iimodeli ezisebenza kakhulu zokubhala ikhowudi, zophando, zokuvelisa umxholo, zee-app ezisebenzisa iintlobo ezahlukeneyo zolwazi, zokwakha ii-arhente, nezokufakwa nokusetyenziswa kwizixhobo zalapha okanye kwizakhiwo zombutho. Amandla ayo aphambili aquka ukufikeleleka kweeparamitha zayo ezivulelekileyo (ngaphandle kokubotshelelwa kumboneleli omnye), izakhono ezibalaseleyo zokuqiqa nokubhala ikhowudi, ukufikeleleka ngeendleko, inkxaso eyomeleleyo yeelwimi ezininzi, kunye nesantya esiphezulu sokuphuculwa okuqhubekayo esivela kwinkqubo yeemveliso neenkonzo zakwa-Alibaba.

Uphuculo lwamva nje luka-2026 (uluhlu lwe-Qwen3.5) luzise inkqubela enkulu kwi-AI esebenzisa ii-arhente, umxholo omde ngakumbi, ukudityaniswa okuphuculweyo kweentlobo ezahlukeneyo zolwazi, kunye nokusebenza okungcono kakhulu ngedola nganye, nto leyo eyenza ukufakwa nokusetyenziswa ngomlinganiselo omkhulu kube nokwenzeka lula ngakumbi.

Abasebenzisi nabaphuhlisi bayayincoma kakhulu impembelelo yayo:

• “I-Qwen3.5-Coder ikhuphisana ngokwenene ne-Claude 3.5 Sonnet kwimisebenzi yokubhala ikhowudi ngoxa ivuleleke ngokupheleleyo kwaye inexabiso eliphantsi kakhulu lokuyisebenzisa” (Injineli ye-AI, e-USA).
• “Sitshintshele lonke ikhonkco lethu lokusebenza kwee-arhente kwi-Qwen — umgangatho wokuqiqa nezakhono zokusebenzisa izixhobo ziyamangalisa ngeli xabiso” (i-CTO, kwishishini le-AI elisaqalayo laseYurophu).
• “Yeyona modeli ivulelekileyo ibalaseleyo endiyisebenzisileyo ngo-2026. Indlela esebenza ngayo kwiintlobo ezahlukeneyo zolwazi nakumxholo omde indimangalise kakhulu” (Umphandi, eAsia).

Yomelele ngokukhethekileyo kumaqela afuna ukusebenza okukwelona nqanaba liphambili ngaphandle kwezithintelo zomthombo ovaliweyo, kwiimeko zokusetyenziswa kwiinkqubo ezisebenzayo apho iindleko zibalulekile, ekwakheni ii-arhente ezilungiselelwe iimfuno ezithile, nakubani na okuxabisayo ukungafihlisi nokulawula ingqokelela yakhe yobuchwepheshe be-AI.

Izinto ezingeloncedo: Nangona yomelele kakhulu, ezona modeli zivaliweyo ziphambili (i-GPT okanye i-Claude yamva nje) zisenokuyodlula kancinane kweminye imisebenzi yokuyila okanye efuna ukuqonda umahluko ofihlakeleyo kakhulu; ukusebenzisa ezona nguqulelo zinkulu (ezinamakhulu eebhiliyoni zeeparamitha) kufuna izixhobo zekhompyutha ezinamandla kakhulu; kwaye, njengakwiimodeli ezininzi ezivulelekileyo, ukufumana ezona ziphumo zibalaseleyo ngamanye amaxesha kufuna ukuyilwa ngononophelo kwemiyalelo nokulungiswa kwemodeli ngoqeqesho olongezelelweyo. Izibonelelo ezifana nokudityaniswa ne-Hugging Face, amaxwebhu asemthethweni, noluntu olusebenzayo lwe-Qwen zibalasele.

Lilonke, kubaphuhlisi nakwimibutho efuna i-AI enamandla, evulelekileyo, efikelelekayo ngeendleko nekumda wophuhliso lobuchwepheshe ngo-2026, i-Qwen inika ukusebenza okubalaseleyo, ukuguquguquka, nexabiso elihle — nto leyo eyenza ibe ngomnye wabadlali ababaluleke kakhulu kwinkqubo yobuchwepheshe be-AI evulelekileyo.

Umthombo onamandla wobukrelekrele obuvulelekileyo obukwelona nqanaba liphambili — fumana nomvuzo we-1 QUA nge-Quasa!  

Iinkwenkwezi eziyi-4.8/5 (ibalasele ekusebenzeni, ekuvulekeni, nasekusebenzeni ngeendleko eziphantsi; amanqaku ehle kancinane ngenxa yezikhewu ezithile ezibakho ngamaxesha athile xa ithelekiswa nezona modeli zivaliweyo ziphambili kwimisebenzi ekhethekileyo kakhulu).

Qalisa: https://quasa.io/projects/qwen