Uyisebenzisa njani i-PC yakho njengeNdawo ye-AI yeNdawo: Isikhokelo esiSebenzayo kunye nokuthelekisa

¿Uyisebenzisa njani iPC yakho njengendawo ye-AI yasekhaya? Ubukrelekrele bokwenziwa ayiseyiyo i-domain ekhethekileyo yeenkampani ezinkulu okanye iingcali zamafu. Abasebenzisi abaninzi nangakumbi bajonge ukonyusa izisombululo ze-AI ngokuthe ngqo kwiikhompyuter zabo zobuqu kwimisebenzi esukela kwisizukulwana sokubhaliweyo ukuya kwi-automation yokuyila okanye iinkqubo zobugcisa, zonke zinobumfihlo obukhulu kwaye ngaphandle kokuxhomekeka kwiiseva zangaphandle. Guqula iPC yakho ibe yindawo ye-AI yasekhaya Yinyani efikelelekayo kwaye ifikeleleka phantse kuye nawuphi na umthandi, ingcaphephe okanye umfundi, nokuba izixhobo zakho azikho kumgangatho wezobugcisa.

Kweli nqaku, uya kufumana indlela yokuguqula ikhompyuter yakho ibe ngundoqo we-AI ecosystem yakho. Siza kujonga ezona ndlela zicetyiswayo zesoftware, iingqwalasela eziphambili malunga nehardware, iimodeli, kunye neempawu, kunye nezibonelelo zokusebenza kunye ne-AI yasekhaya malunga nobumfihlo kunye nobuntu. Ngaphezu koko, ndiya kukukhokela ekukhetheni, ekufakeni, nasekufumaneni okuninzi kwiimodeli zeLLM, ii-apps, kunye nezixhobo, uthelekisa ezona nkqubo zibalaseleyo kwaye ndikunike iingcebiso zokwenza amava akho e-AI agudiswe kwaye akhuseleke, nokuba kukwiWindows, Mac, okanye kwiLinux.

Kutheni usebenzisa iPC yakho njengendawo ye-AI yasekhaya?

Ukusebenzisa ikhompyuter yakho njengeqonga le-AI elisembindini libonelela ngezinto eziluncedo ekunzima ukuzithelekisa neenkonzo zelifu. Esinye sezona zizathu zibalulekileyo kubumfihlo: xa usebenzisana ne-chatbots efini, idatha yakho kunye nezicelo zigcina zigcinwe kwiiseva zomntu wesithathu kwaye, nangona iinkampani ziphumeza amanyathelo okhuseleko, Kuhlala kukho umngcipheko wokuvuza okanye ukusetyenziswa kakubi. Ukucubungula ulwazi ekuhlaleni kuthetha ukuba unolawulo olupheleleyo kwidatha yakho. Akukho mntu wumbi unofikelelo kwimibuzo yakho, iimpendulo, okanye iifayile.

Enye inzuzo enkulu kukungabikho kweemfuno zoqhagamshelwano lwe-Intanethi. Ngenkqubo ekwindawo, ungonwabela iimpawu ze-AI nokuba unonxibelelwano olungazinzanga, uhlala kwindawo enokhuselo olubi, okanye ufuna nje ukusebenza ngaphandle kweintanethi ngezizathu zokhuseleko. Ukongeza, ukwenza ngokwezifiso kukhulu kakhulu: Unokukhetha imodeli ekufanelekela kakhulu, uzenzele ngokweemfuno zakho, kwaye ulungelelanise kakuhle yonke ipharamitha-into enqabileyo enokwenzeka ngeenkonzo zelifu ezinkonkxiweyo.

Akukho nto ibalulekileyo inkalo yezoqoqosho. Nangona iinkonzo zefu zibonelela ngeenguqulelo zamahhala, ukusetyenziswa okuphezulu kubandakanya ukubhaliswa, iintlawulo zamathokheni, okanye ukusetyenziswa kwezixhobo. Xa usebenza ekuhlaleni, ekuphela komda kumthamo wehardware yakho.

Kufuneka wenze ntoni ukuze uqalise? I-Hardware kunye neemfuno ezisisiseko

Umbono oqhelekileyo wokuba ukusebenza ne-AI kufuna iikhompyuter ezinqamlekileyo okanye ii-GPU ezinamandla kakhulu ngoku yinto yexesha elidlulileyo. Imifuziselo yeelwimi yangoku ilungiselelwe ukuba isebenze kwiikhompyuter zasekhaya, kwaye uninzi lwayo, ngakumbi ezo zilinganiselweyo, inokuqhuba nangaphandle kwekhadi lemizobo elizinikeleyo, usebenzisa kuphela i-CPU.

Ukusebenza kakuhle kunye namava amnandi, kuyacetyiswa ukuba ube ne-8-16 GB ye-RAM. kunye neprosesa yangoku enengqiqo (Core i5 okanye i7 yesizukulwana sesithandathu ukuya phambili, okanye iRyzen elingana). Ukuba usebenza ngeemodeli ezinkulu okanye ufuna ukusebenza ngokukhawuleza, i-GPU ene-4GB ye-VRAM yenza umahluko, ngakumbi kwimisebenzi efana nokuveliswa kwemifanekiso okanye iimpendulo ezinde kakhulu zesicatshulwa.

KwiMac, iitshiphusi ze-Apple M1 nangaphezulu zikwaxhasa iimodeli zeLLM zalapha ekhaya ngamaxesha okuphendula amnandi kakhulu. Ngamafutshane, ukuba iPC yakho okanye ilaptop ingaphantsi kweminyaka esixhenxe ubudala, unokuqalisa ukuzama i-AI yendawo.

Zeziphi ii-apps kunye namaqonga owudingayo ukuze ujike iPC yakho ibe yindawo ye-AI yasekhaya?

Intliziyo yenkqubo yakho ye-AI yasekhaya zizicelo ezikhethekileyo ezivala umsantsa phakathi kwezixhobo zakho zehadi kunye neemodeli ze-AI. Phakathi kwezona zinto ziphawuleka ngokulula kwazo, amandla kunye nokuguquguquka, kufanelekile ukukhankanya:

• GPT4All: Enye yezona ndlela zidumileyo kunye nezinobuhlobo. Ikuvumela ukuba ukhuphele kwaye ufake uninzi lweemodeli zolwimi, usebenzisane nazo, kwaye uqwalasele iiparamitha ezahlukeneyo. Yi-cross-platform (iWindows, iMac, kunye neLinux) kwaye inkqubo yofakelo yayo ilula njengayo nayiphi na enye inkqubo yedesktop.
• Jan AI: Igqamile kujongano lwayo lwangoku, ukukwazi ukucwangcisa imisonto yencoko, kunye nokuhambelana kwayo kunye neemodeli zasekhaya kunye nezikude (ukusuka kwi-OpenAI, umzekelo, nge-API). Ukongeza, inikezela nge-API yayo yendawo elinganisa i-OpenAI's, ivumela uJan ukuba adityaniswe njenge-backend ye-AI kwezinye izicelo ezifuna isitshixo se-ChatGPT API, kodwa ngaphandle kokuxhomekeka kwi-Intanethi.
• I-Llama.cpp kunye ne-LM Studio: Ezi zixhobo zikuvumela ukuba usebenzise imifuziselo yeLLM ekuhlaleni kwaye unikeze ufikelelo kwithala leencwadi elibanzi lemifuziselo evela kuBuso obuHugging kunye nezinye iindawo zokugcina.

Inkqubo esisiseko idla ngokulandelayo: Khuphela i-app ekhethiweyo kwiwebhusayithi yayo esemthethweni, yifake kwinkqubo yakho, kwaye ukhangele igalari yeetemplates ezikhoyo (ezihlala zibizwa ngokuba yi "Hub" okanye ezifanayo). Apho ungakhetha imodeli oyifunayo, khangela ubungakanani bayo kunye neemfuno zememori, kwaye ukhuphele yonke into kwi-interface ngokwayo.

Iimodeli eziphezulu ze-AI zokufakela ekuhlaleni

Ihlabathi leemodeli ze-LLM zemithombo evulekileyo likhulu kwaye likhula rhoqo. Ngaphandle kwezo zinikezelwa yi-OpenAI (efuna uxhulumaniso lwamafu), kukho iindlela ezininzi ezilungiselelwe ukusebenza kwindawo: Mistral 7B, TinyLlama Chat, Nous Hermes 2, Mixol 8X 7B, phakathi kwabanye. Uninzi lwezi modeli zibalwa, oku kuthetha ukuba zithatha indawo encinci kwaye zifuna i-RAM encinci ngexabiso lokuncama ixabiso elincinci lokuchaneka.

Kwabaqalayo Iimodeli ezincinci eziphakathi, ezifana ne-Mistro Instruct 7B okanye i-TinyLlama Chat, ziyacetyiswa, njengoko zikhupha ngokukhawuleza kwaye zingalayishi inkqubo. Ukuba ikhomputha yakho ine-RAM eninzi kunye nendawo yokugcina, zama iimodeli ezipheleleyo ezinje ngeMixol 8X 7B, ukwazi ukuba, umzekelo, inokufuna ukuya kuthi ga kwi-26 GB yendawo yedisk nje yemodeli.

Phantse kuzo zonke izicelo ungahluza imifuziselo ngokusekwe kubungakanani bazo, ulwimi lokuqala, iilayisensi okanye uhlobo lwemisebenzi abayiqeqeshelweyo. (ukubhala okubhaliweyo, ukwenziwa kwekhowudi, uguqulo, njl. njl.). Eyona njongo icacileyo yemodeli, kokukhona uchaneka ngakumbi iziphumo ozifumanayo.

Inkqubo yesinyathelo-nge-nyathelo yokufaka kunye nokusetyenziswa kwe-AI yendawo

1. Khuphela kwaye ufake isicelo: Yiya kwiwebhusayithi esemthethweni yesixhobo osithandayo (umzekelo, GPT4All okanye Jan AI), khuphela isifakeli senkqubo yakho yokusebenza, kwaye ulandele amanyathelo akwisikrini. Kwi-Windows, iqhele ukuba yi-wizard yakudala; KwiMac, inokufuna ukwenza iRosetta yeekhompyuter ezineM1/M2 iprosesa; KwiLinux, uya kuba neDEB okanye iiphakheji zeAppImage ezikhoyo.

2. Jonga kwaye ukhuphele iimodeli ze-AI: Nje ukuba uvule usetyenziso, fikelela kumhloli wemodeli (kwi-GPT4All yi "Discovery Model Space", ngoJan AI, "I-Hub"). Hluza, phinda ujonge amanqaku, kwaye xa ufumana imodeli enomtsalane kuwe, cofa u-"Khuphela." Uya kwaziswa ngobungakanani kunye neemfuno phambi kokuba uqhubeke.

3. Ukhetho kunye nokuphunyezwa kokuqala: Nje ukuba ithempleyithi ikhutshelwe, yikhethe kwi-app kwaye uqale incoko entsha okanye umsebenzi. Bhala umbuzo okanye isicelo sakho kwaye ulinde impendulo. Ukuba ubona iimpendulo ezicothayo, zama ukusebenzisa iimodeli ezikhaphukhaphu okanye uhlengahlengise iisetingi.

4. Lungisa iiparamitha kunye nomfuniselo: Kwiinkqubo ezininzi unokuguqula inani eliphezulu lamathokheni (okunciphisa ubude beempendulo), kunye nezinye iinkcukacha ezifana neqondo lokushisa, i-top_p, njl. Zama izicwangciso ezahlukeneyo de ufumane ibhalansi phakathi kwesantya kunye nomgangatho weziphumo ezisebenzela wena.

5. Lungiselela kwaye wenze ngokwezifiso imisonto: Iiprogram ezininzi zikuvumela ukuba udale imicu yengxoxo ngamagama ahlukeneyo kunye neenjongo (iingcamango zevidiyo, ukubhala ngokuyilayo, uncedo ngekhowudi, njl.), kwaye unokugcina imiyalelo yesiko kwintambo nganye, ehambisa intsebenziswano.

Ulawulo lwezibonelelo kunye nokwenza ngcono ukusebenza

Owona mda uphambili we-AI yasekhaya yihardware: Xa imodeli inkulu kakhulu kwi-RAM yakho, ukucotha, ukungqubana, okanye iimpazamo zokwenza. Ii -apps ezilungileyo zibonelela ngezilumkiso kwangaphambili xa ukhetha imodeli enzima kakhulu kwisixhobo sakho.

UJan AI uyagqwesa ngokudibanisa i-on-screen resource monitor ekubonisa ngexesha lokwenyani ukusetyenziswa kwe-RAM, i-CPU kunye nesantya sokusebenza (iithokheni ngomzuzwana). Ngale ndlela, unokuhlala usazi ukuba iqela lakho likumda walo okanye ukuba usenakho ukucudisa ngaphezulu kulo.

Ukuba iPC yakho inekhadi lemizobo leNvidia kwaye ufuna ukulisebenzisa, Ezinye izicelo zivumela ukukhawuleza kweGPU ngokufaka iCUDA. Oku kunokuphinda-phinda isantya kwimisebenzi enzima. Soloko ubhekisa kumaxwebhu asemthethweni ukuze ufake ngokufanelekileyo kwaye uvule inkxaso ye-GPU.

Izinto eziluncedo zokulinganisa: iimodeli ezilula kunye nezisebenzayo

Igama eliqhelekileyo xa uthetha nge-AI yendawo kukuba "quantization." Oku kubandakanya ukunciphisa ukuchaneka kokugcina iintsimbi zemodeli ngokuziguqulela zibe ngamanani anamasuntswana ambalwa, nto leyo enciphisa kakhulu idiski yemodeli kunye nobungakanani bememori, kunye neempembelelo ezincinci kumgangatho wokuphendula.

Uninzi lweemodeli ezikhutshelwayo sele ziza kubalwa kwiinguqulelo ezahlukeneyo (i-4-bit, i-8-bit, njl.). Ukuba imodeli oyifunayo ikhona kuphela kwinguqulo "epheleleyo" kwaye iqela lakho alikwazi ukuyihambisa, kukho izicelo ezikuvumela ukuba uzilinganise ngokwakho (umzekelo, GPTQ).

Obu buchule yenza ukuba kwenzeke iimodeli ezinamandla kwiiPC ezindala okanye ezinobuncwane obulinganiselweyo, ngelixa ugcina ubumfihlo kunye nokuzimela kwilifu.

Ukuthelekisa ezona zixhobo zibalaseleyo ze-AI zendawo: GPT4All vs. Jan AI

Zombini izicelo zibonelela ngayo yonke into oyifunayo ukuguqula iPC yakho ibe lihabhu ye-AI enamandla, kodwa nganye ineempawu zayo ezizodwa ezinokukunceda ukhethe enye okanye enye ngokuxhomekeke kukhetho lwakho.

• Ukusebenziseka ngokulula: GPT4All Kulula kakhulu, ukufakela kuyakhawuleza kwaye ukukhuphela iimodeli kwenziwa kwi-interface ecacileyo kunye nomsebenzisi-friendly. UJan AI, kwelinye icala, unikezela ngombutho wencoko ophucukileyo kunye nokukwazi ukwenza ngokwezifiso imiyalelo kunye nokuhamba komsebenzi.
• Ukuhambelana: Zombini zixhasa iWindows, iMac kunye neLinux. I-Jan AI yongeza ukudibanisa ngokuthe ngqo kunye nezinye izicelo nge-API yendawo.
• Ukujongwa kwezixhobo: I-Jan AI ibonelela ngedeshibhodi yexesha langempela lokusetyenziswa kobutyebi, luncedo kumaqela anemiqobo. I-GPT4All inika ingxelo yobuncinci beemfuno kwaye iyakulumkisa ukuba i-hardware yakho inokusilela.
• Izandiso: UJan ikuvumela ukuba ufakele izandiso ezandisa ukusebenza (umzekelo, i-monitor yesixhobo ekhankanywe ngasentla), engekhoyo kwi-GPT4All.

Ingcebiso yam kukuba uzame zombini kwaye ubone ukuba yeyiphi eyona ifanele ukuhamba komsebenzi wakho kunye neqela lakho.

Ukusombulula iingxaki kunye nee-FAQs

Kuqhelekile ukudibana nemingeni ethile xa ukhuphela kwaye ufaka iimodeli ze-AI, ngakumbi xa ujongana neefayile ezinkulu okanye unezixhobo ezilinganiselweyo kwiqela. Enye yeempazamo eziqhelekileyo kukusilela ukulanda. Kwezi meko, luluvo olulungileyo ukujonga uqhagamshelo lwakho, cacisa indawo yedisk, okanye uqalise kwakhona usetyenziso. Uluntu lwenkxaso yenkqubo nganye, kunye nee-wikis okanye iiforam zabo ezisemthethweni, zihlala zibonelela ngezisombululo zamanyathelo ngamanyathelo.

Ngokubhekiselele kukhuseleko, ukusebenzisa i-AI yendawo kucace ngakumbi kunokunxibelelana neenkonzo ezikude. Idatha yakho kunye nembali yencoko ihlala kwisixhobo sakho kwaye ayisetyenziselwa ukuqeqesha i-algorithms yangaphandle. Nangona kunjalo, njengesilumkiso, kuyacetyiswa ukuba ungabelani ngolwazi olubuthathaka nakwesiphi na isicelo se-AI, nasekhaya.

Kuthekani ukuba ufuna ukusebenza ngakumbi? Ukuba unako ukuhlawula uphuculo lwe-RAM (i-16 okanye i-32 GB) okanye i-GPU yangoku, iimodeli ezinkulu ziya kuhamba kakuhle kwaye uya kukwazi ukuvavanya iimpawu eziphucukileyo ezifana nokusebenzisana kwe-multimodal (umbhalo, umfanekiso, ilizwi). Ngaphandle koko, kukho iimodeli ezikhaphukhaphu, eziphuculwe kakhulu ezisebenza kakuhle kakhulu kwimisebenzi yemihla ngemihla.

Amava akwi-intanethi ngokupheleleyo: Nje ukuba iimodeli zikhutshelwe, isicelo sisebenza ngaphandle koqhagamshelo lwe-Intanethi, ukwandisa ubumfihlo kwaye ikuvumela ukuba usebenze kuyo nayiphi na imeko.

Inkqubo ehlala iphuhla ye-AI yasekhaya

Izisombululo zangoku ze-AI zasekhaya zeePC zifikelele kwinqanaba lokuvuthwa ngoku lizenza zibe yindlela eyomeleleyo kwiinkonzo zelifu. Uluhlu olukhulu lweemodeli, ukufakelwa lula, kunye nesakhono sokwenza ngokwezifiso kukufikelela kwidemokhrasi kubukrelekrele bokwenziwa.

Iinkampani ezifana neGoogle kunye neMicrosoft nazo zinegalelo kwinxalenye yazo ngokusebenzisa amaqonga aphakathi (umzekelo, i-AI Hub okanye iCopilot kwiWindows), kodwa amandla okwenyani e-AI yasekhaya alele kwinto yokuba. Unokwenza ihabhu yakho yesiko ukuba ihambelane nomsebenzi wakho othile, ubumfihlo kunye neenjongo..

Ukwazi ukuba ungumsebenzisi we-AI ocacileyo, sicebisa ukuba uqale ukufunda ngakumbi kwaye uthathe ithuba lobuchule be-ChatGPT kunye nabanye, kuba umzekelo, ngoku uthelekiso lwexabiso kwi-ChatGPT.

Ngoku unazo izixhobo, izikhokelo kunye namaqhinga ayimfuneko okuguqula iPC yakho ibe yindawo yokwenyani yobukrelekrele, Ukuthatha inguqulelo entsha kunye nolawulo olupheleleyo kulwazi lwakho kwelinye inqanaba. Siyathemba ukuba ngoku uyayazi indlela yokusebenzisa iPC yakho njengendawo ye-AI yasekhaya.