Ling-3.0-tiny – imodeli yokucabanga exubile yemvelo eyethulwe yi-Ant Ling Uyasebenza

I-Ling-3.0-tiny iyimodeli yokucabanga exubile yemvelo eyethulwe yi-Ant Ling, enamapharamitha angu-7.9B esewonke futhi esebenzisa angu-1.3B kuphela ithokheni ngayinye, elungiselelwe ukucabanga ngezibalo, ukulandela imiyalelo kanye nokuthunyelwa okuzwelayo ezinsizeni. Imodeli isekela ukubiza amathuluzi ngokwemvelo, futhi ingenza imisebenzi yokulawula i-UI ezenzakalelayo kumadivayisi eselula nasezipheqululini. Itholakala manje ku-OpenRouter naku-Vercel, kanti izisindo zemodeli zizokhishwa njengomthombo ovulekile maduze.

Ukwakheka kokucabanga exubile: Amapharamitha angu-7.9B esewonke, kusetshenziswa angu-1.3B kuphela ithokheni ngayinye, okwenza ukubala kusebenze kahle emisebenzini yokucabanga ngezibalo nokulandela imiyalelo.
Ukubiza amathuluzi ngokwemvelo: Amakhono okusebenzisa amathuluzi akhelwe ngaphakathi, akwazi ukuqondisa ngokuqondile ukulawulwa kwe-UI yesiphequluli kanye nemisebenzi yokuzenzakalela kumadivayisi eselula.
Ukucabanga okwenziwa ngokuphelele endaweni: Kungasebenza ngaphandle kokuncika efwini; ngemva kokuhlanganiswa ne-obsidian-cli, kungasesha, kuhlele futhi kukhiqize okuqukethwe kusizindalwazi sombhalo sasendaweni.
Ukukhiqiza umbhalo okulula: Kunikeza ukukhiqizwa kombhalo okunokubambezeleka okuphansi kwezimo ezisetshenziswa njalo, njengokuhumusha amakhasi ewebhu endaweni nokuphatha imisebenzi yansuku zonke.

Ukwakheka kochwepheshe abaxubile abangenazinsiza (MoE): Inani lamapharamitha emodeli lifinyelela ku-7.9B, kodwa ngokusebenzisa indlela yomzila olawulwa yisango kusetshenziswa cishe angu-1.3B kuphela ithokheni ngayinye. Lokhu kugcina amandla okuveza ulwazi emodeli enkulu, kuyilapho kunciphisa umthamo wokubala wokucabanga ube sezingeni lemodeli encane eminyene.
Indlela yemvelo yokucabanga exubile: Kunemizila eminingi yokucabanga ehlukene eyakhelwe ngaphakathi, ekwazi ukukhetha ngamandla phakathi kwempendulo esheshayo nelula noma indlela yokucabanga ejulile enezinyathelo ezilandelanayo kuye ngobunkimbinkimbi bomsebenzi, ilinganise ukunemba kokucabanga ngezibalo nesivinini sokuphendula ekulandeleni imiyalelo yansuku zonke.
Umzila olawulwa yisango nokulinganisa umthwalo: Kusetshenziswa umzila ofundekayo onikeza ithokheni ngayinye iqoqo lochwepheshe elilungele kakhulu, kanye nomsebenzi wokulahlekelwa wokulinganisa umthwalo ovimbela ukuwa kochwepheshe futhi uqinisekise ukuthi isikhala samapharamitha angu-7.9B sisetshenziswa kahle nangokulinganayo.
Ukushumekwa kwemvelo kokubiza amathuluzi: Ngesikhathi sokuqeqeshwa kwangaphambili noma sokuqondanisa, amaphethini okusebenzisa amathuluzi afakwa ngokuqondile esikhaleni sokumelwa kwemodeli, ukuze imodeli ikwazi ukukhiqiza imiyalelo yokubiza ehambisana nemithetho yesixhumi esibonakalayo samathuluzi ngaphandle kwama-plugin angaphandle, futhi yenze ukulawula okuzenzakalelayo kwe-UI yesiphequluli namadivayisi eselula.
Ukwenza kube lula ukuthunyelwa ohlangothini lwedivayisi: Kusetshenziswa ukucindezelwa ngokulinganisa kanye nokuthuthukiswa kwememori ye-KV-Cache ezimweni zokucabanga kwasendaweni, okusekela ukusebenza ngokuphelele kungaxhunyiwe ku-inthanethi kuhadiwe yabathengi, kususe ukuncika efwini futhi kunciphise izindleko zenethiwekhi nezokubala zokubiza ama-API njalo.

Vakashela inkundla: Vakashela iwebhusayithi esemthethweni ye-OpenRouter (https://openrouter.ai) noma ikhasi le-Vercel AI Gateway.
Khetha imodeli: Sesha bese uvula ikhasi lemininingwane yemodeli ethi inclusionai/ling-3.0-tiny:free.
Thola imvume: Bhalisa noma ungene ku-akhawunti, bese uthola ngokuqondile ukhiye we-API futhi ubize insizakalo ngesikhathi sokuhlolwa kwamahhala.
Linda ukukhishwa komthombo ovulekile: Landela iziteshi ezisemthethweni ze-Ant Ling, bese ulanda amafayela emodeli ukuze uyithumele endaweni ngemva kokukhishwa kwezisindo njengomthombo ovulekile.
Hlanganisa amathuluzi: Lungiselela i-obsidian-cli noma ama-plugin okuzenzakalela kwesiphequluli ukuze unike amandla ukusesha ulwazi lwasendaweni nokulawula i-UI.

Ukukhanya ngokwedlulele: Ingasebenza kusetshenziswa amapharamitha angu-1.3B kuphela, izindleko zokucabanga nokusetshenziswa kwememori yevidiyo kuphansi kakhulu kunamamodeli aminyene anosayizi ofanayo, okuyenza ifaneleke ukuthunyelwa kumadivayisi nasemaphethelweni.
Ubumfihlo kuqala: Isekela ukusebenza ngokuphelele endaweni, idatha ayiphumi kudivayisi, futhi ihlangabezana nezidingo zezimo ezifana nezezimali nezempilo lapho ukuphepha kwedatha kubaluleke kakhulu.
Amathuluzi akhelwe ngaphakathi: Ingabiza amathuluzi angaphandle ngaphandle kokulungiswa okwengeziwe noma ama-plugin, yehlise umkhawulo wokuthuthukisa izinqubo zokuzenzakalela.

Umsizi ohlakaniphile wedivayisi yeselula: Uqeda ukuhumusha, ukuphendula imibuzo nokuphatha uhlelo losuku ungaxhunyiwe ku-inthanethi ocingweni, futhi uphendule ngaphandle kokuxhumeka kunethiwekhi.
Ukuzenzakalela kwesiphequluli: Ngokubiza amathuluzi ngokwemvelo, wenza ngokuzenzakalela ukugcwaliswa kwamafomu ewebhu, ukuqoqwa kolwazi nemisebenzi yokuxhumana ne-UI.
Ukuphathwa kolwazi lwasendaweni: Uhlanganiswa ne-obsidian-cli ukuze kwakhiwe isizindalwazi solwazi esiyimfihlo ngokuphelele, esisekela ukusesha imibhalo yasendaweni nokufingqa okuhlakaniphile.
Ukuthunyelwa emaphethelweni ebhizinisi: Uthunyelwa kumakhompyutha okulawula izimboni noma ezindaweni zenethiwekhi yangaphakathi, unikeze izinsizakalo ze-AI ezithobela imithetho kumadivayisi anezinsiza ezilinganiselwe.
Izinsizakalo ezilula ezisetshenziswa njalo: Usebenza njengesixhumi esibonakalayo esihlala sisebenza ngemuva ukuze uphathe izicelo eziphindaphindwayo ezifana nokuhumusha amakhasi ewebhu nokukhiqiza okuqukethwe, unciphise kakhulu izindleko zokubala.