Uyisebenzisa Kanjani I-PC Yakho Njengehabhu Le-AI Yasendaweni: Umhlahlandlela Osebenzayo Noqhathanisayo

¿Ungayisebenzisa kanjani i-PC yakho njengehabhu le-AI lendawo? Ubuhlakani bokwenziwa akusesona isizinda esikhethekile sezinkampani ezinkulu noma ochwepheshe bamafu. Abasebenzisi abaningi ngokwengeziwe bafuna ukusebenzisa izixazululo ze-AI ngokuqondile kusuka kumakhompuyutha abo ukuze enze imisebenzi esukela ekukhiqizeni umbhalo kuye kokuzenzakalela izinqubo zokudala noma zobuchwepheshe, zonke ezinobumfihlo obukhulu futhi ngaphandle kokuncika kumaseva angaphandle. Guqula i-PC yakho ibe ihabhu le-AI lendawo Kuyiqiniso elithengekayo futhi lingafinyeleleka cishe yinoma yimuphi umshisekeli, uchwepheshe noma umfundi, ngisho noma okokusebenza kwakho kungekona okusezingeni eliphezulu.

Kulesi sihloko, uzothola ukuthi ungayiguqula kanjani ikhompuyutha yakho ibe umnyombo we-AI ecosystem yakho. Sizobheka ezinye izindlela ezinconyiwe zesofthiwe, ukucatshangelwa okubalulekile mayelana nehadiwe, amamodeli, nezici, kanye nezinzuzo zokusebenza ne-AI yendawo ngokokubili ubumfihlo kanye nokwenza kube ngokwakho. Futhi, ngizokuqondisa ekukhetheni, ekufakeni, nasekuzuzeni okuningi kumamodeli we-LLM, izinhlelo zokusebenza, nezinsiza, ngiqhathanise izinhlelo ezingcono kakhulu futhi nginikeze amathiphu okwenza okuhlangenwe nakho kwakho kwe-AI kushelele futhi kuvikeleke, kungakhathaliseki ukuthi uku-Windows, Mac, noma i-Linux.

Kungani usebenzise i-PC yakho njengehabhu le-AI lendawo?

Ukusebenzisa ikhompyutha yakho njengenkundla ye-AI emaphakathi kunikeza izinzuzo okunzima ukuzifanisa nezinsizakalo zamafu. Esinye sezizathu ezibaluleke kakhulu ubumfihlo: lapho usebenzisana nama-chatbot efwini, idatha yakho nezicelo zigcina zigcinwe kumaseva ezinkampani zangaphandle futhi, nakuba izinkampani zisebenzisa izindlela zokuphepha, Kuhlale kunengozi yokuvuza noma ukusetshenziswa kabi. Ukucubungula ulwazi endaweni kusho ukuthi unokulawula okuphelele kudatha yakho. Akekho omunye umuntu okwazi ukufinyelela imibuzo yakho, izimpendulo, noma amafayela.

Enye inzuzo enkulu ukungabikho kwezidingo zokuxhuma ku-inthanethi. Ngohlelo olusendaweni, ungakwazi ukujabulela izici ze-AI ngisho noma unoxhumo olungazinzile, uhlala endaweni enokufakwa okungekuhle, noma umane ufuna ukusebenza ungaxhunyiwe ku-inthanethi ngenxa yezizathu zokuphepha. Ngaphezu kwalokho, ukwenza ngokwezifiso kukhulu kakhulu: Ungakhetha imodeli ekufanele kakhulu, uyenze ngendlela oyifisayo ngokwezidingo zakho, futhi ulungise kahle yonke ipharamitha—into engavamile ukuba yenzeke ngamasevisi wamafu ethini.

Okunye okubalulekile yisici somnotho. Nakuba izinsiza zamafu zinikeza izinguqulo zamahhala, ukusetshenziswa okuthuthukisiwe kufaka phakathi ukubhalisa, izinkokhelo zamathokheni, noma ukusetshenziswa kwezinsiza. Uma usebenza endaweni, umkhawulo kuphela umthamo wehadiwe yakho.

Yini oyidingayo ukuze uqalise? Izingxenyekazi zekhompuyutha nezimfuneko eziyisisekelo

Umbono ojwayelekile wokuthi ukusebenza ne-AI kudinga amakhompiyutha asezingeni eliphezulu noma ama-GPU anamandla kakhulu manje sekuyinto yesikhathi esedlule. Amamodeli wamanje wezilimi alungiselelwe ukusebenza kumakhompuyutha asekhaya, futhi amaningi awo, ikakhulukazi lawo alinganiselwe, ingasebenza nangaphandle kwekhadi lehluzo elizinikele, usebenzisa i-CPU kuphela.

Ukuze usebenze kahle futhi ube nesipiliyoni esijabulisayo, kunconywa ukuthi okungenani ube ne-RAM engu-8-16 GB. kanye nephrosesa yesimanje ngokunengqondo (i-Core i5 noma i-i7 yesizukulwane sesithupha ukuya phambili, noma okulinganayo kwe-Ryzen). Uma usebenza ngamamodeli amakhulu noma ufuna ukusebenza ngokushesha, i-GPU ene-4GB ye-VRAM yenza umehluko, ikakhulukazi emisebenzini efana nokukhiqiza izithombe noma izimpendulo zombhalo ezinde kakhulu.

Ku-Mac, ama-Apple M1 chips nangaphezulu nawo asekela amamodeli wendawo we-LLM anezikhathi zokuphendula ezinhle kakhulu. Ngamafuphi, uma i-PC yakho noma i-laptop yakho ingaphansi kweminyaka eyisikhombisa ubudala, ungaqala ukuhlola nge-AI yendawo.

Yiziphi izinhlelo zokusebenza nezinkundla ozidingayo ukuze uguqule i-PC yakho ibe ihabhu le-AI lendawo?

Inhliziyo yesistimu yakho yasendaweni ye-AI yizinhlelo zokusebenza ezikhethekile ezivala igebe phakathi kwezingxenyekazi zekhompuyutha zakho namamodeli e-AI. Phakathi kokuphawuleka kakhulu ekusebenziseni kwabo kalula, amandla nokuguquguquka, kufanelekile ukusho:

• I-GPT4All: Enye yezinketho ezithandwa kakhulu futhi ezinobungane. Ikuvumela ukuthi ulande futhi ufake inqwaba yamamodeli olimi, uxhumane nawo, futhi ulungiselele amapharamitha ahlukene. I-cross-platform (i-Windows, Mac, ne-Linux) futhi inqubo yokufaka ilula njenganoma yiluphi olunye uhlelo lwedeskithophu.
• Jan AI: Igqama ngesixhumi esibonakalayo sesimanje, ikhono lokuhlela imicu yengxoxo, nokuhambisana kwayo nawo womabili amamodeli asendaweni nakude (kusuka ku-OpenAI, isibonelo, nge-API). Ukwengeza, inikeza i-API yayo yasendaweni elingisa i-OpenAI, evumela u-Jan ukuthi ahlanganiswe njenge-backend ye-AI kwezinye izinhlelo zokusebenza ezidinga ukhiye we-ChatGPT API, kodwa ngaphandle kokuthembela ku-inthanethi.
• I-Llama.cpp ne-LM Studio: Lawa mathuluzi akuvumela ukuthi usebenzise amamodeli e-LLM endaweni futhi unikeze ukufinyelela kulabhulali ebanzi yamamodeli asuka ku-Hugging Face nezinye izinqolobane.

Inqubo eyisisekelo ivamise ukuba kanje: Landa uhlelo lokusebenza olukhethiwe kuwebhusayithi yalo esemthethweni, lufake kusistimu yakho, bese uphequlula igalari yezifanekiso ezitholakalayo (ezivamise ukubizwa ngokuthi “Ihabhu” noma okufanayo). Lapho ungakhetha imodeli oyifunayo, uhlole usayizi wayo nezidingo zememori, bese ulanda yonke into kusixhumi esibonakalayo uqobo.

Amamodeli e-AI aphezulu azofakwa endaweni

Umhlaba wamamodeli we-LLM wemithombo evulekile mkhulu futhi ukhula njalo. Ngaphandle kwalezo ezinikezwa yi-OpenAI (edinga uxhumano lwamafu), kunezindlela eziningi ezilungiselelwe ukusebenza endaweni: i-Mistral 7B, i-TinyLlama Chat, i-Nous Hermes 2, i-Mixol 8X 7B, phakathi kwezinye. Eziningi zalezi zinhlobo zenziwa inani, okusho ukuthi zithatha indawo encane futhi zidinga i-RAM encane ngezindleko zokudela inani elincane lokunemba.

Kwabaqalayo Amamodeli amaphakathi amancane, njenge-Mistro Instruct 7B noma i-TinyLlama Chat, ayanconywa, njengoba ephuma ngokushesha futhi engalayishi kakhulu isistimu. Uma ikhompuyutha yakho ine-RAM eyengeziwe nesikhala sokugcina, zama amamodeli aphelele kakhulu njenge-Mixol 8X 7B, wazi ukuthi, isibonelo, ingadinga kufika ku-26 GB wesikhala sediski kumodeli nje.

Cishe kuzo zonke izinhlelo zokusebenza ungakwazi ukuhlunga amamodeli ngokusekelwe kusayizi wawo, ulimi oluyisisekelo, amalayisensi noma uhlobo lwemisebenzi abayiqeqeshelwe. (ukubhala umbhalo, ukukhiqiza amakhodi, ukuhumusha, njll.). Uma inhloso yemodeli icacisa kakhulu, uthola imiphumela enembe kakhulu.

Isinyathelo ngesinyathelo inqubo yokufaka nokusebenzisa i-AI yendawo

1. Landa futhi ufake uhlelo lokusebenza: Iya kuwebhusayithi esemthethweni yethuluzi olithandayo (isb., GPT4All noma Jan AI), landa isifaki sesistimu yakho yokusebenza, bese ulandela izinyathelo ezikusikrini. Ku-Windows, ngokuvamile kuba wizadi yakudala; Ku-Mac, kungase kudinge ukunika amandla i-Rosetta kumakhompyutha ane-M1/M2 processor; Ku-Linux, uzoba ne-DEB noma amaphakheji we-AppImage atholakalayo.

2. Hlola futhi ulande amamodeli e-AI: Uma usuvule uhlelo lokusebenza, finyelela isihloli samazwe (ku-GPT4All “Isikhala Semodeli Yokutholwa”, ngo-Jan AI, “Ihabhu”). Hlunga, buyekeza izici, futhi lapho uthola imodeli oyithandayo kakhulu, chofoza okuthi "Landa." Uzokwaziswa ngosayizi nezimfuneko ngaphambi kokuqhubeka.

3. Ukukhethwa nokwenza kokuqala: Uma isifanekiso sesilandiwe, sikhethe kuhlelo lokusebenza bese uqale ingxoxo entsha noma umsebenzi. Bhala umbuzo wakho noma isicelo bese ulinda impendulo. Uma ubona izimpendulo ezihamba kancane, zama ukusebenzisa amamodeli alula noma ulungise izilungiselelo.

4. Lungisa amapharamitha nokuhlola: Ezinhlelweni eziningi ungakwazi ukushintsha inani eliphezulu lamathokheni (okukhawulela ubude bezimpendulo), kanye neminye imininingwane efana nezinga lokushisa, top_p, njll. Zama izilungiselelo ezihlukene uze uthole ibhalansi phakathi kwejubane nekhwalithi yemiphumela ekusebenzelayo.

5. Hlela futhi wenze ngendlela oyifisayo imicu: Izinhlelo eziningi zikuvumela ukuthi udale imicu yengxoxo ngamagama nezinjongo ezihlukene (imibono yevidiyo, ukubhala ngobuciko, usizo lokubhala amakhodi, njll.), futhi ungagcina imiyalelo yangokwezifiso yochungechunge ngalunye, olulula ukusebenzisana.

Ukuphathwa kwezinsiza kanye nokwenza kahle

Umkhawulo oyinhloko we-AI yasendaweni yihadiwe: Uma imodeli inkulu kakhulu ku-RAM yakho, ukwehla kokwehla, ukuphahlazeka, noma amaphutha okusebenzisa angenzeka. Izinhlelo zokusebenza ezihamba phambili zinikeza izexwayiso kusengaphambili uma ukhetha imodeli esindayo kakhulu kudivayisi yakho.

U-Jan AI uyaphumelela ngokuhlanganisa isiqaphi sensiza esesikrinini okukubonisa ngesikhathi sangempela ukusetshenziswa kwe-RAM, CPU kanye nesivinini sokucubungula (amathokheni ngomzuzwana). Ngale ndlela, ungakwazi njalo ukwazi uma iqembu lakho lisemkhawulweni walo noma uma usengakwazi ukukhipha okwengeziwe kulo.

Uma i-PC yakho inekhadi lehluzo le-Nvidia futhi ufuna ukusizakala ngalo, Ezinye izinhlelo zokusebenza zivumela i-GPU ukusheshisa ngokufaka i-CUDA. Lokhu kungaphindaphinda isivinini emisebenzini esindayo. Hlala ubheka imibhalo esemthethweni ukuze ufake kahle futhi unike amandla ukusekela kwe-GPU.

Izinzuzo ze-quantification: amamodeli alula futhi asebenza kahle

Igama elivamile uma ukhuluma nge-AI yendawo “quantization.” Lokhu kubandakanya ukunciphisa ukunemba kokugcina izisindo zemodeli ngokuziguqulela izinombolo ezinamabhithi ambalwa, okunciphisa kakhulu idiski yemodeli nosayizi wenkumbulo, okunomthelela omncane kukhwalithi yokuphendula.

Amamodeli amaningi alandekayo asevele ebalwa ngezinguqulo ezihlukahlukene (4-bit, 8-bit, njll.). Uma imodeli oyifunayo ikhona kuphela kunguqulo "egcwele" futhi ithimba lakho lingakwazi ukuyihambisa, kunezinhlelo zokusebenza ezikuvumela ukuthi uzilinganisele wena (ngokwesibonelo, i-GPTQ).

Le ndlela kwenza kube nokwenzeka ukusebenzisa amamodeli anamandla kuma-PC amadala noma anomkhawulo wezinsiza, ngenkathi igcina ubumfihlo nokuzimela ngaphandle kwefu.

Ukuqhathaniswa kwamathuluzi angcono kakhulu endawo e-AI: GPT4All vs. Jan AI

Zombili izinhlelo zokusebenza zinikeza konke okudingayo ukuze uguqule i-PC yakho ibe ihabhu le-AI elinamandla, kodwa ngayinye inezici zayo ezihlukile ezingakusiza ukhethe eyodwa noma enye kuye ngokuthanda kwakho.

• Ukusetshenziswa kalula: GPT4All Kulula kakhulu, ukufakwa kuyashesha futhi ukulanda amamodeli kwenziwa kusuka esibonakalayo esicacile nesisebenziseka kalula. U-Jan AI, ngakolunye uhlangothi, unikeza inhlangano yezingxoxo ethuthuke kakhulu kanye nekhono lokuqhubeka nokwenza ngokwezifiso imiyalelo nokuhamba komsebenzi.
• Ukuhambisana: Zombili zisekela iWindows, Mac neLinux. U-Jan AI wengeza ukuhlanganiswa okuqondile nezinye izinhlelo zokusebenza nge-API yendawo.
• Ukuqapha insiza: U-Jan AI uhlinzeka ngedeshibhodi yesikhathi sangempela yokusetshenziswa kwensiza, ewusizo emaqenjini anezingqinamba. I-GPT4All ibika ubuncane bezidingo futhi ikuxwayise uma ihadiwe yakho ingase ibe mfushane.
• Izandiso: U-Jan ikuvumela ukuthi ufake izandiso ezinweba ukusebenza (isibonelo, isiqaphi sensiza esishiwo ngenhla), esingekho ku-GPT4All.

Isincomo sami ukuthi uzame kokubili futhi ubone ukuthi iyiphi efaneleka kangcono ukuhamba komsebenzi wakho nethimba lakho.

Ukuxazulula izinkinga kanye nama-FAQ

Kuvamile ukubhekana nezinselele ezithile lapho ulanda futhi ufaka amamodeli e-AI, ikakhulukazi lapho usebenza ngamafayela amakhulu noma unezinsiza ezilinganiselwe eqenjini. Elinye lamaphutha ajwayelekile ukwehluleka ukulanda. Kulezi zimo, kuwumqondo omuhle ukuhlola ukuxhumana kwakho, ukusula isikhala sediski, noma uqale kabusha uhlelo lokusebenza. Imiphakathi esekela uhlelo ngalunye, kanye nama-wiki noma izinkundla zabo ezisemthethweni, ngokuvamile bahlinzeka ngezixazululo zesinyathelo nesinyathelo.

Mayelana nokuphepha, ukusebenzisa i-AI yasendaweni kubonakala kakhulu kunokusebenzelana nezinsiza ezikude. Idatha yakho nomlando wengxoxo uhlala kudivayisi yakho futhi akusetshenziselwa ukuqeqesha ama-algorithm angaphandle. Kodwa-ke, njengesexwayiso, kuyanconywa ukuthi ungabelani ngolwazi olubucayi kunoma yiluphi uhlelo lokusebenza lwe-AI, ngisho nasendaweni.

Kuthiwani uma udinga ukusebenza okwengeziwe? Uma ukwazi ukukhokhela ukuthuthukiswa kwe-RAM (16 noma 32 GB) noma i-GPU yesimanje, amamodeli amakhulu azohamba kahle futhi uzokwazi ukuhlola izici ezithuthukile ezifana nokusebenzisana kwe-multimodal (umbhalo, isithombe, izwi). Uma kungenjalo, kukhona amamodeli angasindi, athuthukiswe kakhulu enza kahle kakhulu emisebenzini eminingi yansuku zonke.

Okwenziwayo akuxhumekile ku-inthanethi ngokuphelele: Uma amamodeli eselandiwe, uhlelo lokusebenza lusebenza ngaphandle koxhumano lwe-inthanethi, lwandisa ubumfihlo futhi lukuvumela ukuthi usebenze kunoma yisiphi isimo.

I-ecosystem yendawo ye-AI ehlala ishintsha

Izixazululo zamanje ze-AI zasendaweni zama-PC sezifinyelele izinga lokuvuthwa manje eliwenza abe enye indlela eqinile yezinsizakalo zamafu. Izinhlobonhlobo ezinkulu zamamodeli, ukusebenziseka kalula kokufakwa, namandla okwenza ngokwezifiso afinyelela intando yeningi kubuhlakani bokwenziwa obusezingeni eliphezulu.

Izinkampani ezifana ne-Google ne-Microsoft nazo zifaka isandla ngengxenye yazo ngokusebenzisa izinkundla ezimaphakathi (isb. I-AI Hub noma i-Copilot ku-Windows), kodwa amandla angempela e-AI yendawo alele eqinisweni lokuthi. Ungakwazi ukwenza ihabhu yakho yangokwezifiso ihambisane nokugeleza komsebenzi wakho othize, ubumfihlo, nemigomo..

Ngokwazi ukuthi ungumsebenzisi we-AI ocacile, siphakamisa ukuthi uqale ukufunda okwengeziwe futhi usebenzise amakhono e-ChatGPT namanye, njengoba ngokwesibonelo, manje usungakwazi ukuqhathaniswa kwentengo ku-ChatGPT.

Manje usunawo amathuluzi, imihlahlandlela namasu adingekayo ukuze uguqule i-PC yakho ibe ihabhu leqiniso lobuhlakani bokwenziwa, Ukuyisa ukuqamba okusha kanye nokulawula okuphelele kolwazi lwakho kwelinye izinga. Siyethemba ukuthi manje usuyazi ukuthi uyisebenzisa kanjani i-PC yakho njengehabhu le-AI lendawo.