I-AI yendawo kanye nokwenza izinto ngokuzenzakalela: ama-ejenti, ezokuphepha, kanye namacala angokoqobo

Informatec Digital » Izinsiza » I-AI yendawo kanye nokwenza izinto ngokuzenzakalela: ama-ejenti, ezokuphepha, kanye namacala angokoqobo

La ukuzenzekela nge-AI yendawo Kudlulela ngale kokuba nje kwabathanda ubuchwepheshe abanamaseva asekhaya futhi kuba inketho yangempela kunoma ubani ofuna ukulawula okwengeziwe, ubumfihlo, kanye nokuguquguquka. Namuhla, awusancikile ngokuphelele efwini lenkampani enkulu ukuze ube nama-ejenti akwazi ukufunda isikrini sakho, ukuhambisa igundane lakho, ukusebenza ngamafayela akho, noma ukusebenzisa imisebenzi eyinkimbinkimbi ngemuva.

Isimo siqhumile kakhulu: kusukela izinqwaba ezigcwele njenge-NemoClaw ye-NVIDIA Kusukela kuma-ejenti azimele asebenza kuhadiwe yakho kuya kuzinhlelo zokusebenza zeselula ezifana nePocketBot eziguqula ulimi lwemvelo lube yi-automation yefoni, futhi kufaka phakathi amapulatifomu avulekile njenge-OpenClaw, abasizi abafana ne-Jan AI, kanye neziqondiso ezisebenzayo zokusetha "i-ChatGPT yakho yasekhaya" nge-Ollama kanye ne-Open WebUI, umgomo uyafana: ukwakha uhlelo lwe-ecosystem lapho i-AI ihlala khona kukhompyutha yakho, ixhumana nezinhlelo zakho, futhi izenzele imisebenzi yakho yansuku zonke ngokuzenzakalelayo ngaphandle kokukhipha idatha yakho ohlelweni lwakho.

Kuyini ukuzenzekela kwe-AI yendawo futhi kungani kubalulekile?

Uma sikhuluma ngakho I-AI yendawo yokwenza izinto ngokuzenzakalelaSibhekisela kumamodeli nama-ejenti asebenza kudivayisi yakho (i-PC, iseva, i-DGX, iselula) ngaphandle kokuthumela idatha ebucayi kumaseva angaphandle. Imodeli yenza izinqumo, isebenzisa ikhodi, ifunda amafayela, ibiza ama-API, futhi ixhumanisa amathuluzi, kodwa konke kwenzeka ngaphakathi kwendawo yakho elawulwayo.

Ukuvela kwezinto kube yinto emangalisayo: kusukela kuma-chatbot alula aphendula imibuzo kuphela, sesiqhubekele phambili Ama-ejenti e-AI akwazi ukwenza imisebenzi eminingiukuhlela izinyathelo eziningi, ukubonisana nemithombo yedatha ehlukene, nokwenza izinqumo ezizimele. Lokho kushintshe ngokuphelele indlela esiqonda ngayo ukuzenzekela: imodeli ayisekho nje "ophendulayo," "yilowo othatha isinyathelo."

Lolu shintsho lunomphumela owodwa osobala: Ukuzimela okwengeziwe kusho ingozi eyengeziweUma unikeza i-ejenti ukufinyelela ohlelweni lwamafayela, iziqinisekiso zakho, isiphequluli sakho, noma amathuluzi akho okuthuthukisa, udinga umklamo wokuphepha oqinile. Yilapho izindlela zasendaweni zikhanya khona, ngoba ungakhawulela izimvume, uhlukanise izinqubo, futhi uqaphe eduze lokho okwenziwa yimodeli nganoma yisiphi isikhathi.

Ngaphezu kwalokho, i amamodeli avulekile anamalayisense wamahhala njenge-Apache-2.0 noma i-MIT (Njengezixazululo eziningi ze-Falcon, Bark, Jan, njll.), zikuvumela ukuthi wakhe izixazululo ngaphandle kokuboshwa yizinkontileka noma izinqubomgomo zokusebenzisa ezingacacile. Ungahlola ikhodi, ulungise imodeli, usebenzise ukulungiswa kahle, futhi uyihlanganise nehadiwe ethile efana nama-A100 GPU noma ama-workstation e-NVIDIA DGX.

Kwemikhakha eminingi (ukunakekelwa kwempilo, amabhange, ezomthetho, ezokuphatha umphakathi), lapho Ubumfihlo kanye nesitoreji esiphephile Kungcwele, inhlanganisela I-AI yendawo + ama-ejenti azimele + amamodeli avulekile Kwenza umehluko: uyazenzakalela, kodwa idatha ayisuki emngceleni wakho.

Izitaki ze-AI zasendaweni zokwenza izinto ngokuzenzakalela okuthuthukisiwe: i-NemoClaw, i-OpenShell, ne-OpenClaw

I-NVIDIA ingene kakhulu kulo mdlalo nge I-NemoClawKuyisitaki esivulekile esenzelwe ukufaka ama-ejenti azimele endaweni ngokuphephile nokuqinisekisa ukuthi ahlala evuliwe. Senzelwe ukusebenza emishinini enamandla njenge-NVIDIA DGX Spark, kodwa ifilosofi iyasebenza nakwezinye izindawo eziqinisekisiwe.

I-NemoClaw isebenza njengo capa de orquestación: ifaka futhi ixhumanise i-OpenShell (isikhathi sokusebenza sokuphepha) kanye ne-OpenClaw (uhlaka lwe-ejenti yeziteshi eziningi), ilungiselela ukuqondiswa kwemodeli (nge-Ollama noma i-NVIDIA NIM) futhi isebenzisa izinqubomgomo zokuphepha kusukela ekuqaleni, hhayi njenge-patch yomzuzu wokugcina.

Enhliziyweni yesitaki kuvame ukuba I-NVIDIA Nemotron 3 Super 120BImodeli enamapharamitha ayizigidigidi eziyi-120.000 elungiselelwe ama-ejenti: inekhono kakhulu ekulandeleni imiyalelo eyinkimbinkimbi, amathuluzi okuphatha, kanye nokucabanga okunezinyathelo eziningi. Kodwa-ke, ukuze usebenzise into efana nale, udinga i-GPU engathi sína kanye nenkumbulo eningi; cishe i-87 GB ishiwo kumodeli kuphela.

Isiphetho sivame ukukhonzwa nge I-Ollama njengesikhathi sokusebenza sendawookuveza i-REST API emshinini uqobo. I-NemoClaw ixhumana nale API ukuthumela izixwayiso, ukuthola izimpendulo, nokuhlanganisa izingcingo zamathuluzi kusetshenziswa iphethini yokubiza amathuluzi.

Ingxenye I-OpenShell ibalulekile esicini sokuphephaIsebenzisa i-sandboxing, ilawula iziqinisekiso, isebenza njengommeleli wenethiwekhi, futhi isebenzisa isimiso selungelo elincane kakhulu. Iqapha ukuxhumana okuzanywa yi-ejenti futhi ikuvumela ukuthi uvume noma uvimbele ama-endpoints kusuka ku-interface efana ne-TUI. Ngale ndlela, uma imodeli izama ukufinyelela isevisi entsha, akukho okwenzekayo ngaphandle kwemvume yakho.

Ngaphakathi kwebhokisi lesihlabathi izimpilo I-OpenClaw, ungqimba lwe-ejenti yeziteshi eziningiIphatha ukuxhumana namapulatifomu afana neTelegram, iSlack, neDiscord, iphatha inkumbulo ye-ejenti, ixhumanisa amathuluzi (izikripthi, ama-API, iziphequluli), futhi igcina ingxoxo isikhathi eside. Uma ufuna umsizi ohlala evuliwe, otholakala ngemiyalezo, futhi onenkumbulo eqhubekayo, lena yingxenye eyenza kube nokwenzeka.

Ukuphepha, i-sandboxing, kanye nokuthunyelwa kwendawo isinyathelo ngesinyathelo

Enye yamandla amakhulu alesi sigaba ukuthi Ukuphepha kuyacatshangelwa kusukela esigabeni sokuklamaakungezwanga kamuva. Iphutha elivamile kumaphrojekthi e-ejenti ukuqala ngokwakha zonke izici bese uzama "ukuvikela" lokho osekwakhiwe kakade, okudala imigodi yonke indawo.

Indlela ephakathi yi- i-sandboxing yokwenzaYonke ikhodi i-ejenti efuna ukuyisebenzisa isebenza ngaphakathi kwendawo ehlukanisiwe: ayinakho ukufinyelela okuqondile ohlelweni lwefayela lomsingathi, ayikwazi ukwenza izingcingo zenethiwekhi ezingahleliwe, futhi ayikwazi ukwandisa amalungelo ngaphezu kwalokho okuchazwe ekucushweni.

Lokhu kunciphisa kakhulu umthelela we- ukuhlaselwa okusheshayo komjovo noma imiyalelo enonya. Uma imodeli inquma ukwenza into engavamile, umonakalo uhlala uvalelekile ngaphakathi kwebhokisi lesanti. Noma kunjalo, i-NVIDIA ngokwayo iyavuma ukuthi ayikho ibhokisi lesanti eliphelele, ngakho batusa ukuthi njalo bahlole amathuluzi amasha ezinhlelweni ezihlukanisiwe.

Ngaphezu kwalokho, izinsiza ze-NemoClaw ukulawulwa okuningiliziwe kwamathuluzi nezinqubomgomo ngesikhathi sangempelaNgokuzenzakalelayo, i-ejenti ingaxhumana kuphela nenani elilinganiselwe lama-endpoint enethiwekhi. Uma izama okuthile okusha, i-OpenShell iyayivimba, futhi ungabona kahle ukuthi izama ukwenzani (i-host, i-port, inqubo). Ungabe usuyivuma kuleyo seshini noma wengeze inqubomgomo ehlala njalo ku-host.

Ukugeleza kokufakwa ku-DGX Spark kuvame ukulandela lezi zinyathelo: configure Ubuntu 24.04 LTS enabashayeli be-NVIDIA kulandela umhlahlandlela wokuhlanganisa amakhompyuthaFaka i-Docker 28.xo noma ngaphezulu ngesikhathi sokusebenza se-GPU, faka i-Ollama bese ulanda imodeli ye-Nemotron 3 Super 120B, bese ekugcineni uqalise ukufakwa kwe-NemoClaw ngomyalo owodwa oqala i-wizard yokucushwa.

Lokhu kungeniswa kukuqondisa igama le-sandbox, umhlinzeki we-inference, imodeli ekhethiwe, ukusetha kusengaphambili kokuphepha Futhi, uma ufuna, ukuhlanganiswa kweTelegram. Isikhathi sokusetha esisebenzayo silinganiselwa kumaminithi angu-20-30, kanye neminye imizuzu engu-15-30 yokulanda ithempulethi, kuye ngokuthi i-bandwidth ingakanani.

Ngokuphathelene nokusebenza, kufanele sibe ngokoqobo: impendulo enemodeli yepharamitha engu-120B ingathatha phakathi 30 no-90 imizuzwana endaweni yangakini. Akuyona inkinga ngokwayo, kodwa idinga ukucatshangelwa lapho kuklanywa ukugeleza kokusetshenziswa kanye nohlobo lwemisebenzi oyinika umenzeli.

Ukufinyelela kude, isikhombikubona sewebhu, kanye nehadiwe eyenzelwe i-AI yendawo

Uma konke sekulungisiwe, ungaxhumana ne-ejenti ngezindlela eziningana. Okuvame kakhulu yilokhu ngeTelegramUkusebenzisa i-bot eyenziwe nge-@BotFather, kuyisinqumo esiwusizo: i-API eqinile, ukubethela, izinhlelo zokusebenza zazo zonke izinhlobo zamadivayisi, futhi asikho isidingo sokudalula ama-port eseva yakho ezweni langaphandle.

I-bot ithola imiyalezo yakho, iyithumele ku-ejenti ku-DGX, bese ikuthumelela impendulo. Into ethokozisayo ukuthi, yize ingxoxo idlula engqalasizinda yeTelegram, Ukucabanga kanye nokufinyelela kudatha ebucayi kuhlala kusendaweni eyi-100% emshinini wakho.

Ngaphezu kwalokho, i-NemoClaw inikeza isikhombikubona sewebhu esiyimfihlo Kufinyeleleka nge-URL enethokheni ekhiqizwe kanye kuphela ekupheleni kokufakwa. Kubalulekile ukulondoloza le URL ngokushesha, njengoba ngeke iphinde iboniswe. Ukuze uyibuke komunye umshini kunethiwekhi, kufanele ulungiselele umhubhe we-SSH kanye nokudluliselwa kwe-port usebenzisa i-OpenShell.

Umniningwane owodwa omncane kodwa obalulekile ukuthi i-URL kumele ivulwe nge 127.0.0.1 esikhundleni se-localhostUkusebenzisa i-localhost kungabangela amaphutha e-Unauthorized Origin (CORS), angakuchithela isikhathi uma ungaqapheli.

Kwemisebenzi yansuku zonke kunezindlela eziningana imiyalo ye-CLI ewusizo: vula igobolondo ngaphakathi kwebhokisi lesanti, buka isimo, landela izingodo ngesikhathi sangempela, bhala uhlu lwamabhokisi esanti, qala noma umise ibhuloho leTelegram, yenza kusebenze ukudluliselwa kwembobo, noma usebenzise iskripthi sokukhipha esihlanzekile esisusa yonke inqwaba.

Ngokuqondene nehadiwe, I-NVIDIA DGX Spark Yakhelwe ngokucacile lezi zimo zokusetshenziswa. Luhlelo oluncane olunama-NVIDIA GPU kanye nenkumbulo ehlanganisiwe ye-bandwidth ephezulu, efanelekile ukusebenzisa amamodeli amakhulu namaphakathi ane-latency ephansi ngaphandle kokusetha isikhungo sedatha esigcwele.

La inkumbulo enobunye Kuyasiza ikakhulukazi ngesinye sezithiyo zakudala: ukuhambisa idatha phakathi kwe-CPU ne-GPU. Ngokwabelana ngesikhala sememori, imodeli ifinyelela idatha kahle kakhulu, ivumela amamodeli anamashumi ezigidigidi zamapharamitha ukuthi alayishwe (cishe) ngesikhathi sangempela—okungacabangeki kuze kube muva nje kuhadiwe yabathengi.

Ama-ejenti e-AI endawo adumile: izibonelo kanye namacala okusetshenziswa

Ngaphandle kwe-NVIDIA ecosystem, kunezinhlobo eziningi Ama-ejenti e-AI kanye namapulatifomu agxile ezenzakalelayo eqenjini lakho okufanele ukwazi. Ngayinye ihlose uhlobo oluhlukile lomsebenzisi kanye nesethi ehlukile yemisebenzi.

Isibonelo, i-OpenClaw isithandwa kakhulu njenge ipulatifomu ye-ejenti yomthombo ovulekile esebenza njengomsizi womuntu siqu. Ikuvumela ukuthi udale ama-ejenti ngokwezifiso ukuze ahlanze ibhokisi lakho lokungenayo, athumele imiyalezo, aphathe ikhalenda lakho, ahlele uhambo, noma enze imisebenzi ephindaphindwayo ngokuzenzakalelayo empilweni yakho yedijithali.

Ingafakwa ku- IWindows, iMacOS neLinuxIklanyelwe futhi ukusebenza namamodeli e-LLM endaweni, okuthuthukisa ubumfihlo futhi kunciphisa ukuthembela efwini. Ngaphezu kwalokho, ihlangana nezinhlelo zokusebenza zemiyalezo ezifana ne-WhatsApp, i-Telegram, i-Discord, i-Slack, i-Signal, kanye ne-Apple Messages, ngakho-ke i-ejenti yakho isebenza "ngemuva kwezigcawu" zezingxoxo osuvele uzisebenzisa.

Ngama-plugin, ungayinika ukufinyelela kusiphequluli, amanethiwekhi omphakathi, amaklayenti e-imeyili, nezinye izinhlelo zokusebenza, kanye nokuyivumela xhumana nohlelo lwefayela, sebenzisa imiyalo kanye nezikripthinoma ukwenza imisebenzi ejwayelekile yehhovisi kanye nokukhiqiza ibe ngokuzenzakalelayo. Konke lokhu kugxile ekuvumeleni umsebenzisi ukuthi akhethe ukuthi yimaphi amafolda, izinhlelo zokusebenza, kanye nezinsizakalo ezitholakala kumenzeli.

Kuhlelo olujwayelekile lwe-ecosystem, amapulatifomu afana Ikhompyutha EdidekileLokhu kuguqula i-Perplexity kusuka enjinini yokusesha elula yokuxoxa ibe umsizi okwazi ukwenza imisebenzi eyinkimbinkimbi. Le modi yekhompyutha ikuvumela ukuthi uphequlule iwebhu, udale futhi uphathe amadokhumenti, ubhale ikhodi, ucubungule idatha, futhi uxhumanise nezinsizakalo ezifana ne-Gmail, i-Slack, i-GitHub, kanye ne-Notion.

Amandla ayo asekelwe ekusebenziseni amamodeli afana neClaude, GPT, Gemini, noma i-Sonar kaPerplexity ukuphatha idatha eningi nokuhlukanisa imisebenzi eyinkimbinkimbi ibe imisebenzi engaphansi engenziwa ngokulandelana noma ngokuhambisana. Nakuba kungeyona indawo njalo, iphethini ye-ejenti kanye nokuhlanganiswa namathuluzi kufana kakhulu naleyo yama-ejenti asebenza emshinini wakho.

Embusweni ovulekile kanye nowendawo, Jan AI Ivezwa njenge-ChatGPT ethatha indawo engafakwa ku-Windows, Mac, kanye ne-Linux. Ikuvumela ukuthi usebenzise amamodeli endawo afana ne-Llama (Meta) noma i-Gemma (Google), noma uxhume kumamodeli aku-inthanethi afana ne-ChatGPT, i-Claude, i-Gemini, i-Mistral, i-Qwen, noma i-DeepSeek uma unentshisekelo kungxube.

UJan AI usebenza kokubili njengo umsizi wengxoxo wakudala (buza, bhala phansi, fingqa, humusha, bhala kabusha, chaza) njenge-ejenti ekwazi ukucubungula amafayela namadokhumenti, ukwenza imiyalo, nokukhiqiza ikhodi ngezilimi ezahlukene. Ngaphezu kwalokho, ukugxila kwayo ngokwezifiso kwenza kube lula ukudala i-ejenti yakho enemiyalelo ethile futhi ushintshe phakathi "kwamaphrofayili" ahlukene kuye ngokuthi wenzani.

Ama-ejenti kudivayisi: i-PocketBot kanye nokuzenzakalela kweselula

Umqondo we I-AI yendawo ayihlali ku-PCFuthi kwenza umthelela omkhulu kumafoni aphathwayo, lapho amaphrojekthi amaningi ekhetha amamodeli amancane kodwa akhethekile ukuze enze ifoni isebenze ngokuzenzakalelayo ngaphandle kokudlula efwini.

Isibonelo esicacile yi-PocketBot, i-ejenti esebenza ngqo ku- I-iPhone isebenzisa i-flame.cpp ku-MetalUmgomo wayo ukuguqula ulimi lwemvelo lube yizindlela zokwenza ifoni ngokuzenzakalela: esikhundleni sokuthinta amamenyu noma izinqamuleli eziyinkulungwane, uchaza okufunayo bese i-ejenti inakekela ukukuhumusha kube yizenzo.

I-PocketBot isebenzisa imodeli elinganiselwe ye 3.000 billion amapharamithaIsebenza ngokuphelele endaweni yangakini futhi ngaphandle kokuthumela idatha kumaseva angaphandle. Imemori etholakalayo ku-iPhone 15 Pro ivame ukusetshenziswa ngama-3-4 GB ngaphambi kokuba i-iOS iqale ukubulala izinqubo, ngakho-ke usayizi wemodeli kanye nokulinganisa kubalulekile.

Enye yezinselele abadali bayo abazikhulumayo ukuthola Amamodeli amancane athembekile okubiza amathuluzi kanye nemiphumela ehlelekile ku-JSON. Besebenzisa i-Qwen3, isibonelo, bahlangabezana nezinkinga ezifana namagama epharamitha akhiwe, i-JSON engakhi kahle (amabhuleki angekho) kanye nokunamathela kweskimu okungaguquki, okuphoqa ukusetshenziswa kwezingqimba zokuzilungisa nokuzizama kabusha.

Kukhona futhi impikiswano enkulu mayelana iphuzu elihle kakhulu lokulinganisa Ukuze ufinyelele isilinganiso sekhwalithi/senkumbulo esingcono kakhulu, cabanga ngezinketho ezifana ne-q4_K_M noma i-q5_K_S kuye ngokuthi i-chip ikhiqizwa kanjani kanye nememori etholakalayo. Ingxenye ngayinye encane ekulinganisweni isho amamodeli aphathekayo kakhudlwana, kodwa ingathinta kabi ukucabanga kanye nokunemba kwezingcingo zamathuluzi.

Esinye isici ukulungiswa kwe amapharamitha okusampula kuye ngomsebenzi. Ukucushwa okujwayelekile kufaka phakathi izinga lokushisa elingu-0,7, i-top_p engu-0,8, i-top_k engu-20 kanye ne-repeat_penalty engu-1,1, kodwa kukhona intshisekelo yokuhlukanisa amasu okukhiqiza ingxoxo yamahhala uma kuqhathaniswa nokubiza amathuluzi, lapho ukuzimisela okuningi kanye nobuciko obuncane kuyathakazelisa.

Ekugcineni, efonini ephathekayo ukuphathwa kokuqukethwe Kubucayi kakhulu: isixwayiso sesistimu sivame ukufakwa ku-cache ye-KV ukuze kugwenywe ukucutshungulwa kabusha, futhi amafasitela ashelelayo asetshenziselwa ukugwema ukudlula umthamo; yingakho kuwusizo ukwazi ukuthi kanjani londoloza futhi uhlele izincomo zakho.

Ngaphandle kwalokho, kunendawo yamaqhinga okufingqa okuqhubekayo, inkumbulo ekhethiwe, noma izinhlelo ezihlanganisiwe ezihlanganisa umlando ocindezelwe kanye nomongo oseduze.

Setha "i-ChatGPT yakho yendawo" nge-Ollama kanye ne-Open WebUI

Kulabo abangadingi inqwaba eyinkimbinkimbi njengeNemoClaw, kodwa abayifunayo umsizi wohlobo lwe-ChatGPT osebenza kukhompyutha yakhoIndlela ewusizo kakhulu esekelwe ku-Ollama kanye ne-Open WebUI isithandwa kakhulu.

Umbono ulula: U-Ollama Inesibopho sokulanda nokunikeza amamodeli (Llama, Gemma, Qwen, njll.) emshinini wakho nge-API yendawo, futhi i-Open WebUI inikeza isikhombimsebenzisi sewebhu esifana kakhulu ne-ChatGPT kodwa esisebenza ngokuphelele emshinini wakho. Yonke ithrafikhi phakathi kwe-UI nemodeli idlula i-localhost.

Umhlahlandlela wesinyathelo ngesinyathelo oqondile uchaza ukuthi kanjani, ngezimbalwa Imiyalo engu-15 yokugcinaUngaba nalokhu kusetha nokusebenza ngaphansi kwehora. Kufaka phakathi ukufakwa kwe-Python 3.11, ukucushwa kwesistimu okuyisisekelo, ukufakwa kwe-Ollama, kanye nokufakwa kwe-Open WebUI, kanye nezithombe-skrini namathiphu okuxazulula izinkinga.

Umphumela uba yindawo lapho ujabulela khona akukho zindleko zokubhaliselaUbumfihlo obuphelele (idatha ayiphumi kukhompyutha yakho), izikhathi zokuphendula ezincintisanayo (akukho migqa yeseva eyabiwe), kanye nenkululeko ephelele yokwenza ngezifiso abasizi abakhethekile ukuze bavumelane nezidingo zakho.

Ngaphezu kwalokho, i-Open WebUI ihlanganisa izici ezithuthukisiwe ezifana ne- Ukusesha iwebhu, umhumushi wekhodi, ukudalwa kwemodeli ngokwezifiso Ngokusekelwe ekucushweni okuthile, ilungiselela amakhono e-RAG athuthukile ukwakha izisekelo zolwazi lomuntu siqu. Umqondo uwukuthi ungaba "nomshayeli wendiza oqeqeshiwe" ojwayelene namadokhumenti akho kanye nemisebenzi ngaphandle kokuthembela kubantu besithathu.

Ngemva kwezinyanga ezimbalwa zokusetshenziswa, abasebenzisi abaningi babika ukuthi le nhlanganisela isithathe indawo ngokuphelele [yomkhiqizo/isevisi yangaphambilini]. okubhaliselwe kwabo okukhokhelwayo kuzixazululo zamafungenkathi bethuthukisa ukuhlanganiswa kwedatha yabo yendawo namathuluzi. Isinyathelo esilandelayo esingokwemvelo ukuxhumanisa le "ChatGPT eyenziwe ekhaya" nama-ejenti, izikripthi, kanye nezinsizakalo ukuze kuhlanganiswe ukuzenzekela okuyinkimbinkimbi kakhulu.

Yenza impilo yakho yedijithali isebenze ngokuzenzakalelayo: izibonelo ezisebenzayo nge-AI yendawo

Konke lokhu kuzwakala kukuhle kakhulu ezingeni lobuchwepheshe, kodwa yini ngempela ongayenza empilweni yansuku zonke ngakho? amanxusa endawo aqeqeshwe kahleAmathuba abanzi kakhulu uma uhlanganisa amamodeli e-multimodal, ukufinyelela kwesikrini, amathuluzi, kanye nesitoreji esihlelekile.

Kuneziphakamiso ezenzelwe zenzeka ngokuzenzakalelayo ukusetshenziswa kwekhompyutha yakho nama-ejenti athola izithombe-skrini bese enza okuthile ngazo. Ukugeleza kungaba into efana nale: uhlelo luthatha isithombe-skrini, i-ejenti iyicubungula ngemodeli ekwazi ukusebenza nezithombe, iqonde ukuthi yiluphi uhlelo lokusebenza oluvulekile, yiziphi izinkinobho ezikhona, yimuphi umbhalo ovelayo, futhi ngokusekelwe kusaziso sakho, inquma ukuthi yini okufanele iyenze ngokulandelayo.

Ngale mbono ungenza, isibonelo, setha ama-ejenti okuhumusha akhethekileUhlelo luthwebula ingxenye yesikrini ofuna ukuyihumusha, luyikhulise efasiteleni "lomhumushi wengilazi okhulisayo", futhi lukhiqize ukuhumusha okusheshayo kusetshenziswa imodeli encane (isb., amapharamitha angu-4B) alungiselelwe ukuhumusha, njengenguqulo ye-PHI elungiselelwe kahle.

Esinye isici esithakazelisayo yilesi Amamodeli abonakalayo aguqula izithombe-skrini zibe ama-PDFCabanga ngethuluzi, kusukela kuzithombe-skrini zezethulo, amadeshibhodi, noma amadokhumenti, elikhiqiza ama-PDF ahlelwe kahle ongawalungisa noma uwasebenzise ngqo ezethulweni zakho. Ngokuhlanganisa i-Python ne-Acrobat, ungenza ngokuzenzakalelayo yonke ipayipi.

Ukusebenza ngewebhu ngaphandle kokuthembela kumasevisi angaphandle, ubuchwepheshe basendulo obufana I-BeautifulSoup isawusizo kakhuluUngasetha i-scraper elula ekhasa amakhasi amaningana futhi igcine i-HTML edingekayo kuphela (isibonelo, ukukhipha kuphela