Amandla we-MLOps okulinganisa i-AI kulo lonke ibhizinisi

Ukusho ukuthi kuyinselele ukuzuza i-AI ngezinga lonkana ibhizinisi kungaba ukubukela phansi.

Isilinganiso 54% ku 90% yamamodeli okufunda ngomshini (ML) awakwenzi kube ukukhiqizwa kusukela kubashayeli bezindiza bokuqala ngenxa yezizathu ezisukela kudatha nezindaba ze-algorithm, ukuya ekuchazeni icala lebhizinisi, ukuya ekutholeni ukuthengwa kwesikhulu, kuya ezinseleleni zokushintsha zokuphatha.

Eqinisweni, ukuphromotha imodeli ye-ML ekukhiqizeni kuwumsebenzi obalulekile ngisho nasebhizinisini elithuthuke kakhulu elinabasebenzi be-ML kanye nochwepheshe bezobuhlakani bokwenziwa (AI) kanye nososayensi bedatha.

I-Enterprise DevOps kanye namaqembu e-IT azamile ukulungisa ukuhamba komsebenzi kwe-IT yefa namathuluzi okukhulisa amathuba okuthi imodeli izothuthukiswa ukuze ikhiqize, kodwa ahlangabezane nempumelelo elinganiselwe. Enye yezinselelo eziyinhloko ukuthi abathuthukisi be-ML badinga ukugeleza komsebenzi okusha namathuluzi afaneleka kangcono indlela yabo yokuphindaphinda kumamodeli wokubhala amakhodi, ukuwahlola nokuwaqalisa kabusha.

Amandla we-MLOps

Kulapho MLOps ingena: Isu livele njengeqoqo lezinqubo ezihamba phambili esikhathini esingaphansi kweminyaka eyishumi eyedlule ukubhekana nenye yezindlela izivimbamgwaqo eziyinhloko ukuvimbela ibhizinisi ukuthi lifake i-AI esenzweni – ukuguquka ukusuka ekuthuthukisweni nasekuqeqesheni ukuya ezindaweni zokukhiqiza.

Gartner iyachaza Ama-MLOps njengenqubo ebanzi “ehlose ukwenza lula ukuthuthukiswa kokugcina kuze kube sekugcineni, ukuhlola, ukuqinisekiswa, ukuthunyelwa, ukusebenza kanye nokufakwa kwemodeli ye-ML. Isekela ukukhishwa, ukwenza kusebenze, ukuqapha, ukuhlola nokulandelela ukusebenza, ukuphathwa, ukuphinda kusetshenziswe, ukuvuselela, ukugcinwa, ukulawulwa kwenguqulo, ukuphathwa kwengcuphe nokuhambisana, nokuphathwa kwamamodeli e-ML.”

Ukuphatha amamodeli kwesokudla ukuze uthole isikali

Vnoma i-AI umsunguli kanye no-CEO uManasi Vartak, ophothule e-MIT owayehola abafundi abathole iziqu zobunjiniyela bemishini e-MIT CSAIL ukuze akhe i-ModelDB, udale ngokuhlanganyela inkampani yakhe ukuze kube lula ukulethwa kwemodeli ye-AI kanye ne-ML kuwo wonke amabhizinisi ngesilinganiso.

I-dissertation yakhe, Ingqalasizinda yokuphathwa kwamamodeli kanye nokuxilongwa kwemodeliiphakamisa i-ModelDB, uhlelo lokulandela ukuvela nokusebenza okusekelwe ku-ML.

“Nakuba amathuluzi okuthuthukisa amakhodi alungele ukukhiqiza athuthukiswe kahle, ayakala futhi aqinile, amathuluzi nezinqubo zokuthuthukisa amamodeli e-ML asanda kuzalwa futhi ayabhidlika,” kusho yena. “Phakathi kobunzima bokuphatha izinguqulo zamamodeli, ukubhala kabusha amamodeli ocwaningo okukhiqizwa kanye nokwenza lula ukufakwa kwedatha, ukuthuthukiswa nokusatshalaliswa kwamamodeli alungele ukukhiqiza kuyimpi enkulu yezinkampani ezincane nezinkulu ngokufanayo.”

Izinhlelo zokuphatha amamodeli ziwumgogodla wokwenza ama-MLOps asebenze ngezinga eliphezulu emabhizinisini, echaza, okwandisa amathuba okumodela imizamo yempumelelo. Ukuphindaphinda kwamamodeli kungalahleka kalula, futhi kuyamangaza ukuthi mangaki amabhizinisi angawenzi amamodeli yize enamaqembu amakhulu ochwepheshe be-AI ne-ML kanye nososayensi bedatha kubasebenzi.

Ukuthola uhlelo lokuphatha imodeli enwebekayo endaweni kuwumongo wokukala i-AI kulo lonke ibhizinisi. Abathuthukisi bemodeli ye-AI ne-ML kanye nososayensi bedatha batshela i-VentureBeat ukuthi amandla okuzuza isivuno sezinga le-DevOps kusuka ku-MLOps akhona; inselele ukuphindaphinda amamodeli kanye nokuwaphatha ngempumelelo kakhudlwana, usebenzisa izifundo ezitholwe ekuphindaphindweni ngakunye.

I-VentureBeat ibona isidingo esinamandla engxenyeni yamabhizinisi ahlola ama-MLOps. Lokho kubuka kusekelwa yi-IDC isibikezelo ukuthi u-60% wamabhizinisi azobe esesebenzile ukugeleza komsebenzi wawo we-ML esebenzisa ama-MLOps ngo-2024. Futhi, i-Deloitte ibikezela ukuthi imakethe kwezixazululo ze-MLOps zizokhula zisuka ku-$350 million ngo-2019 ziye ku-$4 billion ngo-2025.

Ukwandisa amandla we-MLOps

Ukusekela ukuthuthukiswa kwe-MLOps ngamathuluzi amasha nokugeleza komsebenzi kubalulekile ekukhuliseni amamodeli ebhizinisini lonke futhi uthole inani lebhizinisi kuwo.

Okokuqala, ukuthuthukisa ukulawulwa kwenguqulo yokuphatha kubalulekile ekukhuleni kwebhizinisi. Amaqembu e-MLOps adinga amasistimu okuphatha amamodeli ukuze ahlanganiswe noma alinganise futhi amboze isiteji samamodeli, ukupakisha, ukuphakela kanye namamodeli asebenza ekukhiqizeni. Okudingekayo izinkundla ezinganikeza ukunwebeka kuyo yonke imijikelezo yokuphila yamamodeli e-ML ngesilinganiso.

Futhi, izinhlangano zidinga inqubo yokusebenza engaguquki yamamodeli. Ukuthi ithimba le-MLOps kanye neyunithi yebhizinisi kusebenza kanjani ndawonye ukuze kusebenze imodeli kuyahlukahluka ngokuya ngecala lokusetshenziswa neqembu, kunciphisa ukuthi mangaki amamodeli inhlangano engawakhuthaza abe ukukhiqizwa. Ukuntuleka kokungaguquguquki kushayela amaqembu e-MLOps ukuthi asebenzise indlela esezingeni eliphakeme kuma-MLOps asebenzisa ukuhlanganiswa okuqhubekayo nokulethwa (CI/CD). Umgomo uwukuthola ukubonakala okukhulu kuwo wonke umjikelezo wempilo wayo yonke imodeli ye-ML ngokuba nenqubo enemininingwane eyengeziwe, engaguquguquki yokusebenza.

Okokugcina, amabhizinisi adinga ukwenza ngokuzenzakalelayo ukulungiswa kwemodeli ukuze anyuse amanani esivuno. Uma ukunakekelwa kwemodeli okuzenzakalelayo kungaba ngaphezulu, kuzosebenza kahle kakhulu yonke inqubo ye-MLOps, futhi kuzoba namathuba aphezulu okuthi imodeli ikwenze ukukhiqizwa. Inkundla ye-MLOps nabathengisi bokuphathwa kwedatha badinga ukusheshisa ukwesekwa kwabo okusekelwe kumuntu ukuze kutholakale izindima ezibanzi ukuze banikeze amakhasimende uhlelo olusebenza kahle kakhulu lokuphatha nokuphatha.

Abathengisi be-MLOps bahlanganisa abahlinzeki be-cloud-platform, izinkundla ze-ML nabathengisi bokuphathwa kwedatha. Abahlinzeki bamafu omphakathi i-AWS, i-Google Cloud ne-Microsoft Azure bonke bahlinzeka ngosekelo lweplathifomu ye-MLOps.

I-DataRobot, i-Dataiku, i-Iguazio, i-Cloudera ne-DataBricks ingabathengisi abahamba phambili abaqhudelana emakethe yokuphathwa kwedatha.

I-LeadCrunch isebenzisa kanjani ukumodela kwe-ML ukuze ishayele amaklayenti amaningi okuhola

Inkampani yokukhiqiza ehamba phambili esekwe emafini I-LeadCrunch isebenzisa i-AI kanye nendlela ye-ML enelungelo lobunikazi ukuze ihlaziye idatha ye-B2B ukuze ihlonze amathemba anamathuba aphezulu okuba amaklayenti amanani aphezulu.

Kodwa-ke, ukubuyekezwa nokubuyekezwa kwemodeli ye-ML bekuhamba kancane, futhi inkampani idinga indlela esebenza kahle kakhulu yokubuyekeza amamodeli njalo ukuze inikeze amakhasimende izincomo ezingcono zethemba. Ithimba lesayensi yedatha le-LeadCrunch lihlala libuyekeza futhi lihlunga amamodeli e-ML, kodwa ngamamodeli angaphansi angu-10-plus nesitaki esihlala sivela, ukuqaliswa bekuhamba kancane. Ukuthunyelwa kwamamodeli amasha kwenzeka kuphela izikhathi ezimbalwa ngonyaka.

Bekuyinselele futhi ukuthola amazwibela okuhlola. Imodeli ngayinye yayiphethwe ngendlela ehlukile, eyayingasebenzi kahle. Ososayensi bedatha babe nobunzima bokuthola umbono ophelele wazo zonke izivivinyo ezenziwayo. Lokhu kuntuleka kokuqonda kuqhubekisele phambili ukubambezela ukuthuthukiswa kwamamodeli amasha.

Ukukhipha nokugcina amamodeli ngokuvamile kwakudinga isikhathi esiningi nomzamo ovela eqenjini lonjiniyela le-LeadCrunch. Kodwa njengenkampani encane, lawa mahora ngokuvamile ayengatholakali. I-LeadCrunch ihlole uchungechunge lwezinkundla ze-MLOps ngenkathi ibona ukuthi zingalulamisa kanjani ukuphathwa kwamamodeli. Ngemva kokusesha okubanzi, bakhethe i-Verta AI ukuze iqondise zonke izigaba zokuthuthukiswa kwemodeli ye-ML, ukuguqulwa, ukukhiqizwa nokugcinwa okuqhubekayo.

I-Verta AI ikhulule ososayensi bedatha be-LeadCrunch ekulandeleleni inguqulo nokugcina amamodeli amaningi ahlelekile. Lokhu kwavumela ososayensi bedatha ukuthi benze ukumodela okuningi kokuhlola. Ngesikhathi sokuthunyelwa kokuqala, i-LeadCrunch nayo yayinamaphuzu ezinhlungu ze-21 okwakudingeka ziqondiswe, ne-Verta AI ixazulula i-20 ngokushesha ngemva kokuqaliswa. Okubaluleke kakhulu, i-Verta AI ikhuphule isivinini sokukhiqiza imodeli ngo-5X futhi yasiza i-LeadCrunch ukufeza ukuthunyelwa okukodwa ngenyanga, ithuthuka isuka kokubili ngonyaka.

Amandla anamandla we-MLOps

Amandla e-MLOps okuletha amamodeli esikalini kanye nesivinini se-DevOps ayisikhuthazo esikhulu samabhizinisi aqhubeka nokutshala imali kule nqubo. Ukuthuthukisa izilinganiso zesivuno semodeli kuqala ngesistimu yokuphatha ethuthukisiwe “engakwazi ukufunda” ekuqeqeshweni kabusha ngakunye kwemodeli.

Kudingeka kube nokumiswa okukhulu kwenqubo yokusebenza, futhi imodeli ye-CI/CD idinga ukusetshenziswa hhayi njengesithiyo, kodwa njengohlaka losekelo lwe-MLOps ukuze kuzuzwe amandla ayo.