- Ukusebenzisa indawo yangasese ye-AI kusetshenziswa i-Ollama kanye nesandiso se-Continue.
- Ukusetshenziswa kwamamodeli akhethekile njenge-Qwen2.5-Coder ukulinganisela amandla engxoxo kanye nesivinini sokuqedela ngokuzenzakalela.
- Ukuqhathanisa phakathi kwe-ecosystem yezixazululo ezisekelwe efwini kanye nobukhosi bedatha kumaseva endawo.
Kulezi zinsuku, cishe wonke umuntu onezinhlelo uye wazama ithuluzi le-AI ukuze abasize. Izixazululo ezifana ne-GitHub Copilot noma i-Cursor zinhle kakhulu ngoba ziphakamisa amabhlogo aphelele ekhodi ngokuphazima kweso. Kodwa-ke, kunemininingwane ebalulekile esingenakuyikhohlwa : ukuze la mathuluzi asebenze, ikhodi yakho ivame ukuya kumaseva angaphandle, okungaba yinkinga enkulu uma usebenza kumaphrojekthi ayimfihlo noma ngaphansi kwezivumelwano eziqinile zobumfihlo.
Izindaba ezinhle ukuthi awusadingi ukuthengisa umphefumulo wakho kudeveli noma ukukhokha okubhaliselwe nyanga zonke ukuze ube nomsizi onamandla. Manje kungenzeka ngokuphelele ukusetha uhlelo lokuxoxa oluzenzakalelayo nolwangasese usebenzisa amathempulethi omthombo ovulekile, ikhadi lehluzo elihle, kanye namathuluzi ambalwa wamahhala. Kulokhu, sizobona ukuthi singakha kanjani umsebenzi oncike kuhadiwe yakho kuphela.
Umshayeli wendiza ohambisana nobukhosi bamamodeli endawo
Uma siqhathanisa isevisi ye-GitHub nesixazululo esenziwe ngokwezifiso, umehluko omkhulu ukulawula. Nakuba i-Copilot kulula kakhulu ukuyifaka futhi ilungele ukusetshenziswa, kusho ukuthi idatha yakho iphuma kunethiwekhi yakho yendawo futhi ukhokha imali uma ungeyena umfundi noma ungagcini amaphrojekthi omthombo ovulekile. Ngakolunye uhlangothi, i-stack yendawo ikunikeza ukulawula okuphelele phezu kwemodeli kanye nobumfihlo obuphelele bekhodi yakho.
Kodwa-ke, masingazenzisi: ukusetha iseva yakho akusho ukuthi konke kumahhala. Kufanele ucabangele izindleko zokuqala zezingxenye kanye nokusetshenziswa kwamandla kokusebenzisa i-GPU. Asihlose ukuphinda zonke izici zenguqulo ye-Microsoft Enterprise, kodwa kunalokho ukumboza izidingo zansuku zonke zonjiniyela ngaphandle kokuthi idatha isuke engqalasizinda yenkampani noma yekhaya.
Ukwakhiwa kobuchwepheshe kanye nezidingo eziyisisekelo
Le nqubo iqondile futhi inengqondo. Ngokuyisisekelo, i-VS Code ithatha umongo wekhodi bese iyithumela kusandiso se-Continue . I-Continue isebenza njengebhuloho noma umlamuleli oxhumana no -Ollama , injini enesibopho sokusebenzisa amamodeli olimi lokuhlela akhethekile emshinini wakho.
Ukuqinisekisa ukuthi lokhu kusebenza kahle, uzodinga i-Visual Studio Code efakiwe kanye neseva esebenzisa i-Ollama. I -GPU yesimanje ene-VRAM eyanele kanye nesikhala sediski samahhala esingu-10-15 GB kuyanconywa kakhulu. Okuhle kunakho konke, asikho isidingo sokubhalisa kunoma iyiphi ipulatifomu noma ukudala ama-akhawunti asolisayo.
Umhlahlandlela wesinyathelo ngesinyathelo: Ukufakwa namamodeli
Ukuze ufeze ukusebenza okuhle kakhulu, icebo akusikho ukusebenzisa imodeli eyodwa kuyo yonke into. Okungcono kakhulu, kufanele uhlukanise imisebenzi yengxoxo nemisebenzi yokuqedela ngokuzenzakalela. Ukuze uthole ukuhlaziywa okujulile kanye nengxoxo, ungasebenzisa i-qwen2.5-coder:7b noma ngisho ne-14b uma une-VRAM engaphezu kwe-12 GB. Lawa mamodeli ahamba kancane kodwa ahlakaniphe kakhulu.
Ukuze uqedele ngokuzenzakalela, okuyikho okuvelayo njengoba uthayipha, sidinga okuthile okusheshayo. Yilapho-ke okungena khona. qwen2.5-coder:1.5bNjengoba ilula, ivumela iziphakamiso ukuthi zigeleze ngaphandle kokulibala. Uma isilandiwe kusetshenziswa umyalo ollama pullSingaqinisekisa ukuthi ziyasebenza ngokusebenzisa uhlelo olulula ollama list Esigungwini.
Uma iseva yakho ye-Ollama isemshinini ohlukile kunalapho uhlela khona, khumbula ukuthi ngokuzenzakalelayo ilalela kuphela ku-localhost. Ukuze ulungise lokhu, udinga ukuhlela ukucushwa kwesevisi ku-Ubuntu usebenzisa i-[method/method]. sudo systemctl edit ollama bese wengeza okuguquguqukayo kwemvelo I-OLLAMA_HOST=0.0.0.0:11434Ngemva kokuqala kabusha isevisi, hlola uxhumano oluvela kwi-PC yakho usebenzisa i-curl ukuqinisekisa ukuthi i-firewall ayivimbi uxhumano.
Qhubeka nezilungiselelo ku-Visual Studio Code
I-Continue cishe iyisandiso esiqine kakhulu sokuxhumanisa amamodeli endawo. Vele usifune emakethe yezandiso, usifake, bese uqala kabusha umhleli. Uma isithonjana sivele kubha eseceleni, iya ku- Local Config bese uhlela ifayela le-config.yaml ukuze ulitshele ukuthi iseva yakho ye-AI ikuphi.
Kuleli fayela, sidinga ukuchaza amamodeli. Ngengxoxo, silungiselela imodeli ye-14B bese siyinika uhlelo oluthile lwe-SystemMessage ukuze luhlale luqondile futhi lugweme izincazelo ezingadingekile. Ukuze siqedele ngokuzenzakalela, sabela imodeli ye-1.5B. Ngokulondoloza izinguquko, i-VS Code manje isiyazi ukuthi iyiphi imodeli okufanele iyibuze isikhathi ngasinye lapho sibhala umugqa wekhodi.
Ukuhlola umsizi wendawo
Ukuqinisekisa ukuthi konke kucishe kuphelele, kungcono ukusebenzisa iskripthi esinamaphutha avamile: iziguquguquko ezingaqalwanga, ukuhlanganiswa kwezintambo okuphelelwe yisikhathi, noma amaphutha okubhala. Ngokukhetha ikhodi bese ucindezela u-Ctrl+L , imodeli enkulu izohlaziya ukukhetha bese iphakamisa ukuthuthukiswa. Indlela elula ukusebenzisa inkinobho ethi Faka esikhundleni sekhodi endala ngentsha ngaphandle kokukopisha nokunamathisela ngesandla.
Ngokuphathelene nokuqedela ngokuzenzakalela, vele uqale ukuthayipha umsebenzi omusha. Ngemva kwesikhashana esifushane, uzobona umbhalo ugqanyiswe ngombala ompunga, ophakamisa okunye komqondo. Uma isiphakamiso silungile, vele ucindezele inkinobho ethi Tab. Uma ubona ukuthi asiphenduli, isixazululo esisheshayo ukuqala kabusha i-extension host kusuka ku-palette yomyalo wonjiniyela.
Ezinye izindlela kanye ne-AI ecosystem
Uma ngesizathu esithile indlela ka-Ollama ingakukhangi, kunezinye izinketho emakethe. I-Amazon ine -CodeWhisperer yayo , enamandla kakhulu, ikakhulukazi kwi-Java ne-Python. Kulabo abafuna okuthile okusekelwe emphakathini, i-Captain Stack iyithuluzi elithakazelisayo elisebenzisa imiphumela ye-Stack Overflow esikhundleni senethiwekhi ye-neural emsulwa.
Siphinde sibe namathuluzi afana ne -Kite , agxile kakhulu ku-Jupyter Lab, noma i-AlphaCode ka-Google , aklanyelwe izinhlelo zokuncintisana. Kukhona nezinketho ezilula kakhulu ezifana ne-Clara Copilot. Kodwa-ke, azikho ezinikeza ubumfihlo kanye nokwenza ngokwezifiso okunikezwa yi-Continue kanye ne-Ollama combo, esebenza ku-GPU yakho.
Okulindelwe kanye nokusebenza okungokoqobo
Kubalulekile ukuthi ungathatheki yi-hype bese uqonda ukuthi imodeli ye-7B noma ye-14B ngeke isuse izintambo zamafu emisebenzini yokwakha eyinkimbinkimbi kakhulu. Ngaphezu kwalokho, ukufaka ama-repository amakhulu kungadla inani elikhulu le-RAM ekuqaleni. Ekugcineni, ikhwalithi yokugcina izohlala incike emandleni ekhadi lakho lehluzo.
Uma une-12GB RTX 3060, ulwazi luzoba luhle kakhulu. Kodwa uma uthuthukela ku- 24GB RTX 4090 , ukusebenza kuba sezingeni lobuchwepheshe, ngezimpendulo zengxoxo ezisheshayo. Kunoma ikuphi, ekuthuthukisweni kwansuku zonke, ukuhlolwa kweyunithi, kanye nokubhalwa kwamadokhumenti, lesi sixazululo singaphezu kokuncintisana.
Njengoba sesihlaziye zonke izinketho, kuyacaca ukuthi singasuka ezinsizeni zokubhalisa siye endaweni esebenza kalula. Kungakhathaliseki ukuthi silungiselela i-Ollama ngamamodeli e-Qwen ukuze sibe nobumfihlo obuphelele noma amathuluzi okuhlola afana ne-CodeWhisperer, i-AI isibe ngakwesokudla somhleli. Ekugcineni, okubalulekile ukuthola ibhalansi phakathi kokusebenza, izindleko, kanye nokuphepha kwedatha ukuze senze ukuhamba komsebenzi kube lula ngangokunokwenzeka.