
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Ababhali: 
 
 
 
 
 
 Thomas Feltin Léo Marchó Juan-Antonio Cordero-Fuertes Frank Brockners Thomas H. Clausen Isishwidi: Ukuchwepheshwa kwe-Deep neural network (DNN) kwidatha yokudlulisa kufuna izinsiza zokubala ukuze kwanelise izidingo zokudlulisela inqubo. Noma kunjalo, izicelo zokufunda ezijulile ezizwelayo nge-latency nokuphepha azikwazi ukuyeka ukuthumela izinsiza zokubala kwi-cloud ekude ngenxa yezindleko zokudlulisa ezibandakanyekayo nokungathembi abanikezi be-cloud bangaphandle. Phakathi kwezixazululo zokukhuphula ukusebenza ngenkathi kugcinwa ukubala endaweni elinganiselwe, ukusheshisa ihadiwe kungaba nzima, futhi ukwenza kahle imodeli kufuna imizamo emikhulu yokwakheka ngenkathi kuvimbela ukunemba. Ukuhlukaniswa kwe-DNN kuyindlela yesithathu eyengeziwe, futhi siqukethe ukusabalalisa umsebenzi wokucubungula phezu kwamadivayisi amaningi we-edge, kucatshangelwa izici zenethiwekhi ye-edge kanye nesakhiwo se-DNN, ngenhloso yokukhuphula ukudlulisela inqubo (inani lokucubungula ngomzuzwana). Leliphepha lethula indlela yokubikezela izikhathi zokucubungula nokudlulisa zokuthunyelwa kwe-DNN okuhlukahlukene, futhi lichaza inqubo yokwenza kahle ukuze kukhuliswe ukudlulisela inqubo. Isixazululi se-branch and bound siye saboniswa futhi sahlaziywa ukuze kulinganiselwe ukusebenza okutholiwe nobunzima. Lokhu kuhlaziya kuye kwaholela ekuchazeni isifunda sokusheshisa, esichaza izimo eziqinisekisiwe ezicini ze-DNN nezinethiwekhi lapho ukuhlukaniswa kwe-DNN kunenzuzo. Ekugcineni, imiphumela yokuhlolwa iqinisekisa izibalo futhi ibonisa ukuthuthukiswa kokudlulisela inqubo ezibonelweni zesampula ze-edge.  Ukwahlukaniswa kwe-DNN ku-edge kuvumela ukusheshisa kokudlulisela inqubo ngenkathi kunciphisa ukusetshenziswa kwebhandwidth kanye ne-latency kumasisekelo aphakathi, isb., kwi-cloud. Isingeniso Amadivayisi axhunyiwe akhiqize cishe ama-bytes angama-2.5 quintillion nsuku zonke ngo-2020.  Kulo mongo, i-Artificial Intelligence (AI) ingenye yezobuchwepheshe ezingabhekana kangcono nenani elikhuphukayo ledatha elikhiqiziwe, ezindaweni eziningi ezifana nokutholwa kwezinto embonini yokubuka, ukuqashelwa kobuso, ukuqashelwa kwezwi, ukucubungula ulimi, noma ukushayela okuzenzakalelayo. I-AI, futhi ikakhulukazi i-Deep Learning (DL), idinga, ngaphezu kwamanani amakhulu edatha, amakhono abalulekile ekusebenzeni kwamandla kanye nesitoreji, okwenza i-cloud computing ibe yisixazululo esaziwa kakhulu sokubamba imisebenzi ye-AI. Ngenxa yalokho, ngo-2019, 96% yemisebenzi ye-AI yenziwa kwi-cloud. 1 2 Izicelo ze-AI zingaba nezikhawulo eziqinile ze-latency, isibonelo, ekushayeleni okuzenzakalelayo, ukuqapha okukhiqizayo, nanoma yikuphi ukucubungula kwesikhathi sangempela okubandakanya ukuxhumana kwezwe langempela. Ngokwesibonelo, izikhathi zokuphendula ezilinganiselayo ekushayeleni okuzenzakalelayo zidingeka ukuba zibe ngaphansi kwemizuzwana eyi-100  , okwenza izikhathi zokubuyela emuva ku-cloud zibe zinde kakhulu kwezicelo ezinjalo. Ngokufanayo, ezinye izicelo zingakhiqiza ukusetshenziswa okubalulekile koxhumaniso kunethiwekhi, isibonelo, esimweni sokucubungula kwevidiyo esezingeni eliphezulu embonini yokubuka, okubavimbela ukuthi baqhume kwi-cloud. Izinqubomgomo zokuphepha kwedatha noma imithetho yokuivikela ingase futhi ivimbele idatha ukuthi ingabuyeli ngaphandle kwezimiso ezithile. Lokhu kukhuthaza ukuthunyelwa kwe-AI ku-edge computing, okubandakanya ukubuyisela imisebenzi yokubala kusuka ezikhungweni zedatha kuseduze namadivayisi e-edge, okungukuthi, eduze kwemithombo yedatha. [](javascript:void()) Uma kuqhathaniswa nama-clouds omphakathi, i-edge iyindawo elinganiselwe ezinsizeni enezikhalazo ezibalulekile  , futhi ukuthuthukisa i-AI kwe-edge ngakho-ke kuyahluka kokuphuma ku-AI ye-cloud. Ukuhlaziya idatha kuseduze nomthombo wayo kungaba nzima ngenxa yamakhono aphansi edivayisi, ukungaguquguquki kwezinsiza ezikhona, kanye nokwehlukahlukana kwezithungathwe ze-edge. Ukunciphisa amandla okubala kusho ukunciphisa ukudlulisela inqubo okutholakalayo, okwenza kube nzima ukubuyisela imisebenzi esindayo kusuka kwi-cloud kuya ku-edge. Kwezinye izicelo, ukudlulisela inqubo, okungukuthi, ukucubungula okutholakalayo ngomzuzwana we-DNN, kungase kuhlobane nokunemba kwe-DNN, isibonelo, ngokuthola izinto emihlubulweni yevidiyo, lapho ukudiliza izinga lokucubungula kusuka ku-15 kuye ku-5 ukucubungula ngomzuzwana kuye kwanciphisa isikolo se-F1__3__ somtholi wezinto ngo-10%  . Ngamanye amazwi, ukuthunyelwa okunokudlulisela okungaphansi kwezinhlangano kungadala ukulahleka kolwazi olubalulekile, isibonelo, ukulahleka kokutholakala ezicelo zokugadwa kwevidiyo. [](javascript:void()) [](javascript:void()) Zikhona izindlela ezimbili eziyinhloko ezitholakalayo ukuze kwanelise izidingo zokusebenza ezithungeni zemvelo ze-edge: ukucindezela imodeli noma ukusheshisa ihadiwe.   kuhlanganisa ukunciphisa nokwenza kahle iziinethiwekhi ze-neural ukuze zibe yinguqulo elula, ngokuvamile engenzi kahle, yemodeli. Izindlela ezijwayelekile zokucindezela imodeli zifaka ukususa  , okungukuthi, ukususa amapharamitha angadingekile kumodeli, i-quantization  , okungukuthi, ukunciphisa isitoreji esinikeziwe sokugcina imodeli, kanye nokuncibilikisa ulwazi  , okungukuthi, ukuqeqesha inethiwekhi ye-neural encane ngolwazi oluthathwe kwelinye elikhulu. Ukwenza kahle imodeli kufuna imizamo emikhulu yokwakheka futhi kungaba isithiyo ekunembeni kwemodeli.   kumadivayisi e-edge kusho ukuthola ihadiwe elingaba lihle ekusetshenzisweni kwamandla futhi libe lula, ngenkathi likwazi ukucubungula ukucubungula ngokusebenza okwanele. Zikhona izinhlobo eziningana zihadiwe ezicini zokusheshisa ku-edge, ezihlukahlukene ekusebenzeni kahle nasekubaluleni, kodwa amayunithi okubala ajwayelekile awahlangabezani nalokho okulindelwe kanti ihadiwe elizinikele njenge-ASICs litholakele lingasebenzi futhi libiza kakhulu  . Ukucindezela imodeli [](javascript:void()) [](javascript:void()) [](javascript:void()) Ukusheshisa ihadiwe [](javascript:void()) Ukuhlukaniswa kwe-DNN kuyindlela eyengeziwe yokusheshisa inqubo ngokusebenzisa ubuningi bamadivayisi akhona ezithungeni ze-edge ukusabalalisa ukubala kokucubungula - okungasetshenziswa yodwa, noma ngokubambisana nezinye izindlela ezimbili. Ukuhlukaniswa kwe-DNN kuqukethe ukucatshangelwa kwenethiwekhi ye-neural njengomshini womshini wokuhlukanisa izigaba, nokusabalalisa lezi zigaba kumadivayisi e-edge. Ukubekwa kwezi zigaba kusekelwe kokubili ku-DNN kanye nezici zenethiwekhi engaphansi. Ukuhlukaniswa kwe-DNN kuncike ekuhlonzeni amaphuzu okwahlukana, okuyizindawo kumshini wokubala lapho imodeli ihlukaniswe yaba izigaba. Phakathi nokusebenza, izigaba zisebenza ngokulandelana, ngayinye ithumela imiphumela yokucubungula ephakathi kwesigaba esilandelayo. Lokhu kuvumela isigaba ngasinye ukuthi siqale ukubala idatha elandelayo ngenkathi amanye amadivayisi eqhubeka nokucubungula leyo edluliselwe, ngaleyo ndlela kuthuthukiswa ukudlulisela inqubo, njengoba kuboniswe ku  . Ekupheleni kweliphepha, ukuhlukaniswa kwe-DNN kuzochazwa ngokusebenzisa isibonelo sokucubungula kwesikhathi sangempela kumifudlana yevidiyo. umfanekiso 1 A. Umsebenzi Ohlobene Izindlela zokuhlukaniswa kwe-DNN, ezivela ekudluliselweni okuphakathi kwe-cloud, zifuna ukwenza kahle izinkomba ezahlukahlukene zokusebenza, njenge-latency yokubala, ukusetshenziswa kwamandla, ukusetshenziswa kwezinsiza, izindleko, noma ukudlulisela. Ukuhlukaniswa kwe-DNN kuncike ekuhlotshanisweni kwezinkomba ezilodwa noma ezimbalwa zalezi zinkomba ukuze kuchazwe umgomo wokwenza kahle, nendlela esebenzisa le nkombisi ukuthola izinhlelo zokuhlukaniswa. Ngokwesibonelo, i-Neurosurgeon   ifuna iphuzu elilodwa lokwahlukana, igcina isigaba sokuqala ku-edge futhi idlulisele isigaba sesibili ku-cloud, ukuze kuncishiswe i-latency nokusetshenziswa kwamandla. Izicelo ku-IoT ziye zakhipha ukuhlukaniswa okuhlangene nokudlulisela kwama-DNN amaningi ukuze kwenziwe kahle amandla, ukubambezeleka, kanye/noma izindleko, kusetshenziswa izixazululi ezahlukahlukene njenge-SPSO-GA   elifanisa i-Particle Swarm Optimization (PSO) ne-Genetics Algorithm (GA), noma i-DDPQN   elisebenzisa i-Deep Reinforcement Learning ukuthola izinhlelo zokuhlukaniswa. Ezinye izibonelo zifaka i-DINA  , echaza inkinga ye-Integer Non Linear Programming (INLP), futhi isebenzisa isixazululi esisekelwe kumqondo wokufanisa, ukuze kwenziwe kahle ukubambezeleka nokusetshenziswa kwezinsiza ezithungeni ze-fog. Izindlela ezikhuphula ukudlulisela inqubo zifaka i-DNN surgery  , esebenzisa usayizi ophansi wemiphumela ye-DNN eyengamele ukuhlukanisa ukubala kokucubungula phakathi kwe-edge ne-cloud. Zombili lezi zicelo zenza izigaba ezimbili, enye ku-edge nenye ku-cloud. Ezinye izifundo emkhakheni we-mobile edge-cloud   zifaka futhi ukubala kwe-multi-threaded ku-cloud ukuze kusheshiswe kakhulu ukudlulisela okuphelele. Ukuncika kwi-cloud ekubalweni kokucubungula kwesigaba sesibili kucabanga ukuthi kunokuxhumana kahle kwenethiwekhi, okungase kungabi nesiqiniseko ngoxhumano olungalungile lwezinye izicelo ze-edge ezingahlukanisiwe. I-Edgent   ilungele indlela yangaphambili kumadivayisi eselula namaseva e-edge atholakalayo, kunciphisa isibopho sokudlulisela ku-cloud. Le ndlela ihlotshaniswa nokuqeqeshwa okukhethekile kanye nesakhiwo semodeli, futhi isagcina ukuhlukaniswa kube izigaba ezimbili. Ezinye izindlela ziye zacatshangelwa izikhawulo eziningi zokwahlukana, isibonelo, i-IONN  , echaza inkinga yokuhlukaniswa ukuze kuncishiswe i-latency nokusetshenziswa kwamandla, futhi ibuye icatshangelwe isikhathi sokulayisha amamodeli kumaseva e-edge. [](javascript:void()) [](javascript:void()) [](javascript:void()) [](javascript:void()) [](javascript:void()) [](javascript:void()) [](javascript:void()) [](javascript:void()) Lezi zindlela zombili zihlangana nama-clouds akude, noma zilinganiselwe endleleni yokuhlukaniswa, isibonelo, kudingeka ukuthi zihlukaniswe ngezingxenye ezimbili, nge-multi-threading noma ngaphandle kwayo. Akukho neyodwa yezifundo zokuhlukaniswa kwe-DNN ezibalwe lapha ezicatshangelayo izikhawulo eziningi zokwahlukana kanye ne-multi-threading. B. Izimpahla Leliphepha lethula uhlaka lokucubungula olwahlukene ukuze kukhuliswe ukudlulisela inqubo yokubala kwe-DNN yesikhathi sangempela kumidiya yokudlulisa, ngezikhawulo eziningi zokwahlukana kanye nokuhlukaniswa kwe-multi-threaded. Izimpahla zaleliphepha yilezi: 
 
 
 
 
 
 Imodeli yokubala nokudlulisa izikhathi zezinethiwekhi ze-DNN ezihlukene, lapho kunganqunywa khona ukudlulisela inqubo okungalindelekile kohlelo lokuhlukaniswa. Ukwakheka kwenkinga yokwenza kahle yokuhlukaniswa kwe-DNN, ukuthunyelwa kwesixazululi se-branch and bound salenkinga, nokulinganisa ubunzima bayo. Izibalo ukuze kuhlanganiswe izithuthukisi zokusebenza ezingenzeka ngezici zenethiwekhi eziguquguqukayo nezobu-DNN. Ukuhlonishwa kwezifunda eziqinisekisiwe ezicini zenethiwekhi nezobu-DNN ezihola ekuphumeleleni kwezikhombisa zokuhlukaniswa kanye nezindleko zokubala izixazululo ezinjalo, okungukuthi, izimo lapho ukuhlukaniswa kwe-DNN kunenzuzo. Imiphumela yokuhlolwa echaza izifunda ezichazwe ezibalweni, kanye nokunemba kokubikezela nokuthuthukiswa kokugcina kokudlulisela inqubo ezindaweni ezifanayo nezihlukene. Ngokulandelayo, ukuze kugwenywe ukudideka, i-deep neural network ehlukanisiwe izobhekwa njenge  , futhi igama elithi   lizobhekisela kunethiwekhi yezokuxhumana emphysical. DNN inethiwekhi C. Uhlaka Lwephepha Isiqephu esisele saleliphepha sihlelwe kanje. Isiqephu II sichaza isizinda se-Deep Learning nendlela ama-DNN amelelwe kulesi sifundo. Isiqephu III sichaza izindlela zokubala nokudlulisa izikhathi ezidingekayo ku-Isiqephu IV esichaza inkinga yokwenza kahle, kanye ne-algorithm ye-branch and bound yokwenza izinqumo zokuhlukaniswa. Isiqephu V sichaza imiphumela yezibalo ukuze kulinganiswe ukusebenza kanye nobunzima be-algorithm ye-branch and bound, kanye nezimo ezihola ekuphumeleleni kwenethiwekhi efanayo. Isiqephu VI sichaza imiphumela yokuhlolwa eqinisekisa izibalo ze-Isiqephu V, futhi ibonisa ukuthuthukiswa kokudlulisela inqubo kusethi yezokuhlola ezihlukene. Ekugcineni, Isiqephu VII sixoxa ngemiphumela futhi sihlanganise ukubaluleka okubanzi kwalendlela. ISIQEPHU II. Isizinda: Ukufunda Okujulile Ngaphakathi kwenkundla ye-Machine Learning, i-Deep Learning (DL) ibhekisela ezindleleni ezincikile ekusetshenzisweni kwama-DNN, okungukuthi, iziinethiwekhi ze-neural ezokwenziwa (noma izindlela zokufunda ezihlobene) eziqukethe ungqimba olulodwa kuphela olufihliwe. Okunye kwezinkomba zokuqala ngezakhiwo zokufunda ezijulile kwashicilelwa ngo-1967  , kodwa kwaze kwaba ngama-2010s lapho i-DL yazuza udumo, kuqala ekuqashelweni kwezwi, bese kuthi kamuva ekutholeni izinto ezithombeni, ngenxa yohlelo lokubamba ubudlelwano obuyinkimbinkimbi kudatha enezimiso eziphakeme. I-Deep Learning kulindeleke ukuthi ibambe iningi lezicelo ze-AI ngo-2022, kanti inhlangano engaphezu kuka-75% isebenzisa ama-DNN ezicelo ezingasebenzisa izindlela zakudala. Izindlela ze-DL ziye zachazwa njengo “   ”  . Lezi zimiso zezibalo zivame ukuhlelwa zaba izigaba, ezakha izinyathelo ekubalweni. Igama elithi   libhekisela ezinhlelweni eziqukethe ungqimba olulodwa kuphela. [](javascript:void()) ubuchwepheshe bokufunda komshini lapho izikhalazo zithatha ifomu lezinhlelo zezibalo eziyinkimbinkimbi ngezinga lokuxhumana elilungisekayo [](javascript:void()) deep Zikhona izinhlobo ezimbili eziyinhloko ze-DNN: iziinethiwekhi ze-neural ezidlulisela phambili kanye nezinethiwekhi ze-neural ezihlehlisayo. Zombili izinhlobo ze-DNN zingamelelwa njengezithombe zokubala, lapho umehluko omkhulu ukuthi iziinethiwekhi ezidlulisela phambili zingamelelwa njengezithombe eziqonde ngqo ezingenazo izimbuzana, ngenkathi iziinethiwekhi ezihlehlisayo zingaba nezimbuzana. Iziinethiwekhi ezihlehlisayo zakhelwe izinhlobo zemidiya, lapho into ngayinye yomidiya inokuxhumana okungenzeka kanti izinto zangaphambili emgqeni, ngenkathi iziinethiwekhi ezidlulisela phambili zingacabangi ukuxhumana phakathi kwezinto emgqeni wemidiya. Isiqephu esisele saleliphepha sizocubungula imbangela yeziinethiwekhi ze-DNN ezidlulisela phambili. 5 ISIQEPHU III. Imodeli Yokucubungula Ehlukene Lesi Siqephu sethula imodeli yokumelwa kwezici ze-DNN nezinethiwekhi ukuze kubikezelwe izikhathi zokubala, izikhathi zokudlulisa, kanye nokudlulisela kokugcina inqubo kwanoma yiluphi uhlelo lokuhlukaniswa. A. Ukumelwa Kwe-DNN Nezinethiwekhi I-DNN yokudlulisa phambili enemiphumela engu-N imelelwa njengomshini we-DAG GA=(L,E) o­ne­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­

Part of HackerNoon's growing list of open-source research papers, promoting free access to academic material.

Read My Stories

Cisco Systems

Le audio iveliswe ngolwimi lokuqala lwebali!

Kutheni i-AI Yakho Ingenaso iFu ukuze iSebenze ngokukhawuleza Ngoku

About Author

IZIMVO

ZIJONGE IIMPAWU

ELI NQAKU LINIKEZELWE KU

Related Stories

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps