Ama-DevOps ane-AI kanye ne-LLMOps: kusukela kuphayiphi kuya kumodeli ekhulumayo

Informatec Digital » Izinsiza » Ama-DevOps ane-AI kanye ne-LLMOps: kusukela kuphayiphi kuya kumodeli ekhulumayo

Ukufika kwe-AI ekhiqizayo kanye namamodeli ezilimi ezinkulu kushintshe ngokuphelele indlela isofthiwe eyakhiwe ngayo, esetshenziswa ngayo, futhi esetshenziswa ngayo. Akusanele ukuba namapayipi amahle e-DevOps noma ukusebenzisa ama-MLOps akudala : uma wethula i-LLM ku-equation, ungena endaweni lapho imodeli ikhuluma khona, icabanga, ithuthukisa, futhi ngezinye izikhathi iziphatha ngezindlela ezingalindelekile.

Kulesi simo esisha, amaqembu adinga ukuhlanganisa ama-DevOps, i-AI, nama-LLMOps ukuze aphathe yonke inkathi yokuphila kwezinhlelo zokusebenza ezisekelwe ku-LLM , kusukela ekuhlolweni nasekuthuthukisweni okusheshayo kuya ekufakweni, ukuqapha, ukuphepha, kanye nokwenza ngcono izindleko. Lesi sihloko sichaza ubunzima futhi sikuqondise, isinyathelo ngesinyathelo, ukuthi ungahlanganisa kanjani ama-ChatOps, ama-DevOps, ama-MLOps, ama-GenAIOps, nama-LLMOps ekusebenzeni kwesimanje.

Kusukela ku-DevOps kanye ne-MLOps kuya ku-LLMOps: kungani imodeli ingasasebenzi

Sekuyiminyaka eminingi, amaqembu obunjiniyela ebeka phambili ukulethwa kwesofthiwe ngokuzenzakalela nokunciphisa ukungqubuzana phakathi kwentuthuko nengqalasizinda . Lokhu kwaholela ekuzalweni kwe-DevOps: ukuhlanganiswa okuqhubekayo, ukuthunyelwa okuqhubekayo, ingqalasizinda njengekhodi, ukubonwa, kanye nesiko lokubambisana elaqeda ukudluliselwa okungapheli phakathi kweminyango.

Lapho idatha iba yingxenye yomkhiqizo, ama-MLOp avele njengempendulo yesidingo sokuphindaphindeka kanye nokulandeleka kwamamodeli okufunda komshini . Imikhuba efana nokuguqulwa kwedatha, ukuqeqeshwa kokuhlelwa kwamapayipi, ukutholwa kokukhukhuleka, kanye nokuhlolwa okuqhubekayo kwamamodeli okubikezela kwabekwa ezingeni elifanayo.

Inkinga ukuthi ama-LLM aphula okuningi kokuqagela okucatshangelwe ku-DevOps kanye ne-MLOps . Akuwona ama-API angaguquki noma imisebenzi elula ebuyisela inombolo eqondile: aphendula ngolimi lwemvelo, axube umongo, imiyalelo, amathuluzi, kanye nedatha yesikhathi sangempela, futhi angakhiqiza imiphumela emibili ehlukene yokufaka okufanayo.

Lokhu kusho ukuthi akwanele ukuguqula imodeli kanye nezisindo zayo ; kuyadingeka futhi ukulawula izikhuthazo, amathempulethi, izinqubomgomo zokuphepha ezichazayo, imikhawulo, amathuluzi axhunyiwe, umongo ofakiwe, ngisho nemithetho yebhizinisi ebeka isimo sokuziphatha kwesistimu.

Iyini i-LLMOps futhi iyixazulula kanjani ngempela?

Singabuka ama-LLMOp njengohlaka lokusebenza oluvumela ukuthunyelwa okuphephile, okulawulwayo, nokuqhubekayo, ukugcinwa, kanye nokulinganiswa kwezinhlelo zokusebenza ezisekelwe ku-LLM . Kuyisambulela lapho i-DevOps isebenza khona, ama-MLOp, kanye namakhono amasha akhethekile kumamodeli okukhiqiza ahlala khona.

Empeleni, ama-LLMOp agxila kakhulu "ekuqeqesheni imodeli ephelele" kodwa agxila kakhulu ekulawuleni ukuziphatha kwawo ekukhiqizeni . Lokhu kufaka phakathi indlela ukugeleza okusheshayo okuklanywa ngayo futhi kuhunyushwe ngayo, indlela ama-LLM axhumana ngayo nemithombo yedatha yangaphakathi, indlela izindleko zamathokheni kanye nokubambezeleka okubhekwa ngayo, kanye nendlela ingozi yokusho okuthile (ukungaqondi kahle, ukuvuza kolwazi, ukucwasa, izimpendulo ezinobuthi, njll.) ephathwa ngayo.

Izidingo ezixazululwa yi-LLMOps, kanye ne-DevOps/MLOps zodwa ezingenakukwazi ukuzimboza, zifaka phakathi izici ezahlukahlukene njengokulandelelwa kwengxoxo, ukuhlolwa okuzenzakalelayo kwekhwalithi yempendulo, kanye nokuhlolwa kwe-A/B kwezinhlobo zokuziphatha . Asikhulumi nje ngokunemba okuvamile, kodwa futhi nokuvumelana, ukuvumelana nebhizinisi, kanye nokuphepha.

Ngaphezu kwalokho, izindleko azisekho kuphela ekuqeqeshweni kwamamodeli kanye nokusingatha : isikhuthazo ngasinye, umongo ngamunye owandisiwe, kanye nocingo ngalunye oluhambisanayo kudala ukusetshenziswa kwe-GPU noma amathokheni kuma-API ezentengiselwano. Ngaphandle kwesendlalelo se-LLMOps sokwenza lokhu kusetshenziswa kwemali kubonakale futhi kuxhunywe kumishini, izinsizakalo, kanye nezimo zokusetshenziswa, umthethosivivinywa ukhula ngendlela engalindelekile.

Ama-ChatOps + ama-LLMOps + ama-DevOps: imisebenzi iba yingxoxo

Enye yezindlela ezinamandla kakhulu ukuhlanganiswa kwe-ChatOps kanye ne-LLMOps ngaphakathi kwesiko le-DevOps . Esikhundleni sokukhawulelwa kumadeshibhodi, izikripthi, kanye namapayipi, amaqembu aqala ukusebenzisa ingxenye enkulu yesistimu ngeziteshi zengxoxo ezifana ne-Slack, i-Microsoft Teams, noma i-Discord.

I-ChatOps iphakamisa ukuthi imisebenzi yansuku zonke (ukuthunyelwa, imibuzo yelogi, ukuqalisa kabusha, izinguquko zokucushwa) yenziwe ngama-bots ngaphakathi kwesiteshi sokuxhumana uqobo , ngokusobala eqenjini lonke. Yonke imiyalo, isenzo, kanye nomphumela kuqoshwa engxoxweni.

Uma i-LLM ingezwa kuleyo ndlela, kuvela ungqimba olusha lobuhlakani: ama-chatbot aqonda ulimi lwemvelo, ahumushe izinhloso, futhi angabangela imiyalo eyinkimbinkimbi noma ahlaziye izimo ngaphandle kokuthi umqhubi adinge ukukhumbula yonke imibhalo noma ifulegi eliqondile.

Izibonelo ezijwayelekile zalokhu kuhlangana zifaka phakathi i-bot, enikwe amandla yi-LLM, efunda amamethrikhi avela ku-Prometheus kanye namalogi avela ku-Loki lapho othile ethayipha "insizakalo yeqembu X ihamba kancane" futhi iphakamisa izenzo ezifana nokulinganisa amakhophi, ukwenza i-rollback, noma ukuqalisa izivivinyo ezithile, konke kuchazwe ngolimi lwemvelo.

Ezingeni lamasiko nokusebenza, lokhu kuholela ezinqumweni ezisheshayo, ukungenelela okuncane ngesandla emisebenzini ephindaphindwayo, kanye nokuhlangenwe nakho okubushelelezi kwamaqembu e-DevOps , aqala ngokucima umlilo njalo aye ekusebenzeni ekuthuthukisweni kwamasu.

Izimiso ezibalulekile zomjikelezo wokuphila we-LLM ekukhiqizeni

Ukuqhuba uhlelo lwe-LLM olubalulekile akuyona iphrojekthi eyenzeka kanye kuphela; kungumjikelezo ophindaphindayo lapho ushintsho ngalunye lungashintsha khona ukuziphatha kohlelo . Nakuba inhlangano ngayinye iluvumelanisa nesimo sayo, ngokuvamile kunezigaba eziyisithupha ezinkulu, eziziqinisayo.

Isigaba sokuqala ukuqeqeshwa noma ukuzivumelanisa nezimo kwemodeli , okungasukela ekusebenziseni imodeli eyisisekelo kanye nokusebenzisa ukulungisa kahle, i-LoRa, noma amanye amasu okulungisa ngedatha yakho. Lapha, into ebalulekile akuyona nje ukusebenza, kodwa futhi nokushiya irekhodi eliphelele: amasethi edatha, izihlungi ezisetshenzisiwe, ama-hyperparameter, izinguqulo ze-tokenizer, izakhiwo ezivivinyiwe, njll.

Uma lesi sigaba singenziwa ngendlela ehlelekile futhi singabhalwanga phansi, imodeli izalwa ngaphandle kokubusa . Ngemva kwalokho, cishe ngeke kwenzeke ukuchaza ukuthi kungani isabela ngendlela esabela ngayo noma ukuphinda umphumela othize uma kudingeka ekuhlolweni kwamabhuku.

Isigaba sesibili ukuthunyelwa, lapho imodeli ishiya khona ilebhu bese ingena ekukhiqizweni. Kuma-LLMOp, lokhu akukhona nje "ukuyifaka esitsheni": kufanele unqume ukuthi iyiphi ihadiwe ozoyisebenzisa , ukuthi ungayiphatha kanjani imemori yezimo ezisebenza isikhathi eside, ukuthi iyiphi i-topology yeqembu okufanele isetshenziswe, nokuthi ungayikhulisa kanjani ngokwethrafikhi ngaphandle kokubambezeleka okukhulu noma izindleko zingalawuleki.

Kusukela lapho, ukuqapha okuqhubekayo okugxile ekuziphatheni kuyaqala ukusebenza . Akwanele ukubheka nje ukusetshenziswa kwe-CPU ne-RAM; kuyadingeka ukuqapha ikhwalithi yezimpendulo ezinengqondo, ukuzinza kwesitayela, izinga lamaphutha, ukuvela kwezindleko ngethokheni ngayinye, ukubonakala kwezimpendulo eziyingozi noma ezingahambisani, kanye nezinguquko ezikhathini zokuphendula ngaphansi kwamaphethini okusetshenziswa ahlukene.

Ezigabeni zakamuva, imisebenzi yokwenza ngcono kanye nokulungisa kahle iyenziwa: ukulungisa izixwayiso, ukulungisa i-RAG, ukuhlola izinhlobo zamamodeli, ukulinganisa, ukwenza ukuhlolwa kwe-A/B, ukushintsha izinqubomgomo zokuphepha kwencazelo, noma ukulungisa imithetho yebhizinisi . Kuyinqubo ecishe ibe yobuciko, lapho idatha, ubunjiniyela, kanye nebhizinisi kuhlangana khona ukuze kunqunywe ngezinto eziza kuqala.

Okokugcina, konke lokhu kuhlelwe ngaphakathi kwezendlalelo zokuphepha nokuphatha (ukulawula ukufinyelela, ukuhlola, ukuvimbela ukuvuza, imikhawulo yokusetshenziswa, ukuthobela imithetho) kanye nangaphakathi komqondo wokubuyekezwa okuqhubekayo, lapho imodeli kanye nesistimu yayo yemvelo kulungiswa khona ukuze kuhambisane nezinguquko kudatha, imithethonqubo, kanye nezidingo zangaphakathi.

I-GenAIOps kanye nendlela yokugeleza kwesaziso e-Azure

Ngaphakathi kwendawo yonke ye-LLMOps, kuneziphakamiso eziqondile kakhulu zokuhlela lo mjikelezo wokuphila. Enye yezindawo ezithuthuke kakhulu endaweni yebhizinisi yi -GenAIOps ene-flows esheshayo ku-Azure Machine Learning ehlanganiswe ne-Azure DevOps , enikeza indlela ehlelekile kakhulu yokwakha izinhlelo zokusebenza ezisekelwe ku-LLM.

Ukugeleza okusheshayo akusiyona nje i-prompt editor; kuyipulatifomu ephelele yokuklama, ukuhlola, ukwenza izinguqulo, kanye nokusebenzisa ukugeleza kokuxhumana kwe-LLM , kusukela ezimweni ezilula (i-prompt eyodwa) kuya ekuhleleni okuyinkimbinkimbi okunama-node amaningi, amathuluzi angaphandle, izilawuli, kanye nokuhlola okuzenzakalelayo.

Isici esibalulekile yindawo yokugcina imisebenzi ephakathi nendawo , esebenza njengomtapo wolwazi wenkampani. Esikhundleni sokuba iqembu ngalinye libe nezicelo zalo kumadokhumenti ahlukene noma ezindaweni zalo zokugcina, zihlanganiswa zibe yindawo yokugcina eyodwa ephethwe, enamagatsha acacile, izibuyekezo, kanye nomlando.

Ngaphezu kwalokho, ipulatifomu inezela amakhono okuhlola okuhlukile kanye ne-hyperparameter: kungenzeka ukuhlola inhlanganisela ehlukene yezinkomba, amamodeli, izilungiselelo zokushisa noma izinqubomgomo zokuphepha kuma-node amaningi okugeleza bese uqhathanisa imiphumela nama-metric acacile.

Ngokuphathelene nokufakwa, i-GenAIOps enokuhamba kwesaziso ikhiqiza izithombe ze-Docker ezihlanganisa kokubili ukugeleza kanye neseshini yenqubo , ezilungele ukusebenza ezindaweni ezifana ne-Azure App Services, i-Kubernetes, noma izinqubo eziphethwe. Kusukela kulesi sisekelo, ukufakwa kwe-A/B kuvunyelwe ukuqhathanisa izinguqulo zokugeleza ezindaweni zangempela.

Enye indlela enamandla ukuphathwa kobudlelwano phakathi kwamasethi edatha kanye nokugeleza. Ukugeleza ngakunye kokuhlola kungasebenza namasethi edatha amaningana ajwayelekile kanye nokuhlola , okuvumela ukuqinisekiswa kokuziphatha ezimweni ezahlukene ngaphambi kokubeka okuthile ezandleni zabasebenzisi bokugcina.

Ipulatifomu futhi ibhalisa ngokuzenzakalelayo izinguqulo ezintsha zamasethi edatha kanye nokugeleza kuphela uma kunezinguquko zangempela, futhi ikhiqiza imibiko ephelele ngamafomethi afana ne-CSV ne-HTML ukusekela izinqumo eziqhutshwa idatha, hhayi ukuqonda.

Izigaba ezine ze-GenAIOps ngokugeleza kwesaziso

Indlela ye-GenAIOps ihlukanisa umjikelezo wokuphila ube yizigaba ezine ezihlukaniswe ngokucacile, okusiza ukugwema isiphithiphithi esivamile sokuthi "sizame izinto nge-AI bese sibona ukuthi kwenzekani".

Isigaba sokuqala, ukuqaliswa, sigxile ekuchazeni ngokunembile inhloso yebhizinisi kanye nokuqoqa izibonelo zedatha ezimele . Lapha, isakhiwo esiyisisekelo sokugeleza okusheshayo sichazwe, futhi ukwakheka kuklanywe, okuzobe sekuhlungwa.

Esigabeni sokuhlola, ukugeleza kusetshenziswa kule datha yesampula, futhi kuhlolwa ukuhlukahluka okuhlukene kweziphakamiso, amamodeli, kanye nokucushwa . Ukuphindaphinda kuyaqhubeka kuze kutholakale inhlanganisela eyamukelekayo ehlangabezana nezidingo zekhwalithi ephansi kanye nokuvumelana.

Okulandelayo kuza isigaba sokuhlola nokulungisa, lapho kusetshenziswa khona amasethi edatha amakhulu nahlukahlukene kakhulu ukuze kulinganiswe kahle . Kuphela uma ukugeleza kubonisa ukusebenza okuhambisanayo okuhambisana nezindinganiso ezichaziwe lapho kubhekwa khona ukuthi kukulungele isinyathelo esilandelayo.

Okokugcina, esigabeni sokuqaliswa, ukuhamba komsebenzi kulungiselelwe ukusebenza kahle futhi kusetshenziswe ekukhiqizeni, okuhlanganisa ukuhlolwa kwe-A/B, ukuqapha, impendulo yomsebenzisi, kanye nemijikelezo yokuthuthuka eqhubekayo . Akukho lutho oluqedwa ngempela: ukuhamba komsebenzi kuyaqhubeka nokulungiswa ngokusekelwe ekubonweni kokusetshenziswa kwangempela.

Le ndlela ihlanganiswe kuthempulethi yokugcina ye-GenAIOps, enekhodi yokuqala, amapayipi akhiwe ngaphambilini, kanye namathuluzi okusebenzisa endawo nawefu ukuthuthukisa, ukuhlola, nokuqalisa izinhlelo zokusebenza ezisekelwe ku-LLM ngaphandle kokuvuselela isondo lephrojekthi ngayinye.

Ukuhlanganiswa ne-Azure DevOps: izindawo zokugcina, amapayipi, kanye nokuqinisekiswa

Ukuletha i-GenAIOps kusuka ku-theory iye enhlanganweni yangempela, ukuyihlanganisa ne-Azure DevOps kubalulekile. Ithempulethi evamile iqala ngendawo yokugcina izinto ku-Azure Repos enamagatsha amabili amakhulu, i-main kanye ne-development , ekhombisa izindawo ezahlukene kanye namasu okukhuthaza ikhodi.

Indawo yokugcina isampula iqoqwe kusuka ku-GitHub, ehlotshaniswa ne-Azure Repos, futhi ukuhamba komsebenzi okuvamile kuhilela ukudala amagatsha ezici kusuka kusikhombisi sokuthuthukiswa . Izinguquko zibe sezisunduzwa ngezicelo zokudonsa, eziqala ngokuzenzakalelayo ukuqinisekiswa kanye namapayipi okuhlola.

Ukuze i-Azure DevOps ikwazi ukusebenzisana ne-Azure Machine Learning nezinye izinsizakalo, kuhlelwe inhloko yesevisi ku-Azure njengobunikazi bobuchwepheshe . Lobu bunikazi busetshenziswa ekuxhumekeni kwesevisi ye-Azure DevOps, okuvumela amapayipi ukuthi aqinisekise ngaphandle kokuveza okhiye abacacile.

Ngokuvamile, le nhlangano inezimvume zomnikazi ku-ML subscription noma isisetshenziswa sesisebenzi, okuvumela amapayipi ukuthi ahlinzeke ngama-endpoints, abhalise amamodeli, futhi abuyekeze izinqubomgomo ezitolo ezinkulu . Ukuze uthole ukuphepha okwengeziwe, ingashintshwa ibe yindima yoMnikeli ngokushintsha izinyathelo ze-YAML ezisingatha izimvume.

Ngaphezu kwalokho, kudalwa iqembu lezinto eziguquguqukayo ku-Azure DevOps eligcina idatha ebucayi njengegama lokuxhumeka kwesevisi noma izihlonzi zezinsiza . Lezi zinto eziguquguqukayo zivezwa kumapayipi njengendawo, okugwema isidingo sokufaka ikhodi eqinile yolwazi olubalulekile kukhodi.

Ukuhlela izindawo zokugcina zasendaweni nezikude kuvumela igatsha lokuthuthukisa ukuthi livikelwe ngezinqubomgomo zegatsha ezidinga ipayipi lesicelo sokudonsa ukuze lisebenze ngaphambi kokuhlanganiswa. Leli payipi liphatha ukuqinisekiswa kokwakha kanye nokugeleza kokuhlola, okuvimbela izinguquko eziphukile ukuthi zingeniswe.

Uma ikhodi isingenile ekuthuthukisweni, kuqalwa ipayipi lokuthuthukiswa elihlanganisa izigaba ze-CI kanye ne-CD eziphelele : ukuqhuba izivivinyo nokuhlola, ukubhalisa ukugeleza kurejista yemodeli ye-Azure ML, ukusebenzisa ama-endpoints kanye nokuhlolwa komusi, kanye nokuhlanganiswa kuma-endpoints asanda kudalwa.

Iphethini efanayo iphindaphindwa enguqulweni noma egatsheni lokukhishwa, elixhunywe ezindaweni zokukhiqiza. Lapho, amapayipi e-CI/CD okukhiqiza aphinda umjikelezo wokuhlola, ukuhlola, kanye nokusabalalisa , kodwa engqalasizinda yezinga lokukhiqiza kanye nedatha, ngokulawula okukhulu kanye nokubuyekezwa okwengeziwe ngesandla uma kudingeka.

Isici esibalulekile "ukubuyekezwa komuntu okujikelezayo" okufakiwe kulezi zipayipi: ngemuva kwesigaba se-CI, i-CD ivinjiwe kuze kube yilapho umuntu evuma mathupha ukuqhubeka kusuka ku-interface ye-Azure Pipelines. Uma kungavunywanga phakathi nesikhathi esithile (isibonelo, imizuzu engama-60), ukuqaliswa kuyenqatshwa.

Ukuqaliswa kwendawo kanye nokuxhumana nabahlinzeki be-LLM

Akuyona yonke into eyenzekayo ngamapayipi: I-GenAIOps iphinde isekele ukusebenza kwendawo ukuze kwenziwe ukuhlolwa okusheshayo . Ungakopisha indawo yokugcina ithempulethi, udale ifayela le-.env kufolda yempande, futhi uchaze ukuxhumana ne-Azure OpenAI noma amanye ama-endpoint ahambisanayo ngaphakathi kwayo.

Lokhu kuxhumana kufaka phakathi amapharamitha afana ne-api_key, i-api_base, i-api_type, kanye ne-api_version, futhi kubhekiselwa kuwo ngamagama ngaphakathi kokugeleza (isibonelo, uxhumano olubizwa ngokuthi "aoai" olunenguqulo ethile ye-API). Lokhu kuvumela ukugeleza okufanayo ukuthi kusebenze endaweni kanye nasefwini ngaphandle kokushintsha ikhodi.

Ukuze usebenzise le ndlela, mane nje udale indawo ebonakalayo noma i-conda bese ufaka ukuncika okudingekayo (i-promptflow, i-promptflow-tools, i-promptflow-sdk, i-openai, i-jinja2, i-python-dotenv, njll.). Ukusuka lapho, ungabhala izikripthi zokuhlola kufolda yokusebenza yendawo bese usebenzisa izivivinyo ku-flows echaziwe.

Lokhu kubili kwefu/esakhiweni kuhambisana kahle nomqondo ovuthiwe we-DevOps: ukuhlolwa okuncane kwenziwa endaweni, ukuqinisekiswa okusemthethweni kwenziwa ngamapayipi, bese kuthunyelwa ezindaweni ezinkulu ezinezilawuli nokuhlolwa . Konke kuhunyushwe ku-Git futhi kuxhunywe ku-Azure DevOps.

Amathuluzi ajwayelekile ohlelweni lwe-DevOps olune-AI kanye ne-LLMOps

Ngale kwesiphakamiso esithile se-Azure, uhlelo lwesimanje lwe-DevOps olune-AI kanye ne-LLMOps luvame ukuthembela kusethi yamathuluzi ahlanganisa i-ChatOps, ukuhlelwa kwemodeli, ukuqapha, kanye nokubuka.

Esigabeni se-ChatOps, kuvamile ukuhlanganisa i-Slack nama-bot afana ne-Hubot , i-Microsoft Teams nama-ejenti asekelwe kuma-Power Virtual Agents, noma i-Discord kanye nezinhlaka ezifana ne-Botpress noma i-Rasa ukwakha abasizi abangokwezifiso abaxhumanisa namapayipi, izinhlelo zokuqapha, kanye nezinsizakalo zangaphakathi.

Ensimini ye-LLMOps/MLOps, amapulatifomu afana ne-Kubeflow kanye ne-MLflow avamile ekuphatheni amapayipi, amarekhodi amamodeli kanye nokuhlolwa, ngaphezu kwamathuluzi athile afana ne-Weights & Biases (W&B) okulandelela okuthuthukisiwe kwe-metric, ukuqhathanisa kokugijima noma ukubonwa okujulile.

Ukwakha izinhlelo zokusebenza ku-LLM kuvame ukuhilela ukusebenzisa izinhlaka ezifana ne-LangChain noma imitapo yolwazi efana ne-OpenLLM , ezenza kube lula ukuhlanganisa amaketanga okulandelela, izixhumi kudatha yangaphandle, amathuluzi, kanye nama-ejenti anezinyathelo eziningi. Ngesikhathi esifanayo, kuvela izixazululo zokubonwa kwe-LLM okuqondile, okuvumela ukuqapha izimpendulo, izimpendulo, izindleko, kanye nekhwalithi.

Ngokuhlanganiswa ne-DevOps yakudala, amathuluzi afana ne-Jenkins noma i-GitLab CI ahlala efanele ingxenye ye-CI/CD, i-Kubernetes kanye ne-ArgoCD yokufakwa okuqhubekayo kwe-cloud-native , kanye nama-observability stacks afana ne-Prometheus, i-Grafana kanye ne-Loki kuma-metric, amadeshibhodi kanye nama-log.

Izinselele, imikhawulo kanye nokwamukelwa okuqhubekayo

Konke lokhu kusetshenziswa kwemikhuba namathuluzi akuveli ngaphandle kwentengo. Ubunzima bokuphatha izikhuthazo, izinguqulo zamamodeli, kanye nezinhlobo zomsebenzi bukhulu kakhulu, ikakhulukazi lapho amaqembu amaningi esebenza ngesikhathi esisodwa — isimo lapho amasu afana ne-GitOps ewusizo ekuhlanganiseni izinguquko kanye nokusetshenziswa.

Ngaphezu kwalokho, ama-bot nama-LLM e-ChatOps anamakhono asebenzayo aletha izingozi ezinkulu zokuphepha uma enezimvume eziningi ezindaweni zokukhiqiza noma uma izindawo zokudalulwa kwedatha zingalawulwa kahle.

Okungeziwe kulokhu ukuthembela kumamodeli omthombo ovulekile anamalayisense abucayi noma ama-API ezentengiselwano angashintsha imigomo, amanani, noma imikhawulo. Futhi, okwenza izinto zibe zimbi kakhulu, ukuhlolwa okuqinile kwama-LLM ekukhiqizweni kusalokhu kuyindawo evulekile, kunemibuzo eminingi engakaphendulwa.

Ngakho-ke, kunengqondo ukusondela ekwamukelweni kwe-LLMOps kanye ne-ChatOps ngaphakathi kwe-DevOps ngendlela eqhubekayo nelawulwayo , uqala ngokwenza imisebenzi ephindaphindwayo ngokuzenzakalela ngama-bots alula (ukuqalisa kabusha, imibuzo yelogi, ukumaka kokwakha, njll.).

Kamuva, ama-LLM angafakwa ukuze asekele imisebenzi, ahlukanise izigameko, noma asize ekulungiseni amaphutha , isibonelo, ukuchaza amaphutha avela kumarekhodi noma ukuphakamisa izindlela zokunciphisa ngokusekelwe kumadokhumenti angaphakathi.

Uma ukusebenza kwe-ML yakudala sekuzinzile, sekuyisikhathi sokubhekana nama-LLMOp ngamamodeli olimi akhethekile ezizindeni ezifana nesevisi yamakhasimende, i-DevSecOps noma i-QA, sisebenzisa konke okufundwe ezigabeni zangaphambilini.

Indawo lapho yonke le mikhuba ikhomba khona yindawo yobunjiniyela exoxayo, ebikezelayo futhi ekhula ngokuzimela , lapho iningi lentuthuko kanye nokusebenza kuvezwa ngolimi lwemvelo futhi i-AI isiza ukwenza izinqumo ezisebenzayo mayelana nokusetshenziswa, ukukhuliswa noma ukubuyiselwa emuva.

Njengoba le nkinga ikhona—i-DevOps, i-ChatOps, i-MLOps, i-GenAIOps kanye ne-LLMOps—izinhlangano zinohlaka oluqinile lokwakha nokusekela izinhlelo ezisekelwe ku-LLM eziletha ngempela inani , zigcina ukulawula ikhwalithi, izindleko, ukuphepha kanye nokuhambisana nebhizinisi, esikhundleni sokuhlala kuma-prototypes noma izivivinyo ezihlukanisiwe eziwa ngokushesha nje lapho zifika ekukhiqizweni.