Eminyakeni engaphezu kwamashumi amathathu edlule, iRed Hat yabona amandla okuthuthukiswa komthombo ovulekile kanye nelayisense yokudala isofthiwe engcono nokukhuthaza ukusungulwa kwe-IT. Imigqa eyizigidi ezingamashumi amathathu yekhodi kamuva, iLinux ayizange nje ithuthuke yaba isofthiwe yomthombo ovulekile ephumelela kakhulu, kodwa futhi igcina leso simo kuze kube namuhla. Ukuzibophezela ezimisweni zomthombo ovulekile kuyaqhubeka, hhayi kuphela kumodeli yebhizinisi lenkampani, kodwa futhi njengengxenye yesiko lomsebenzi. Ekuhlolweni kwenkampani, le mibono inomthelela ofanayo kubuhlakani bokwenziwa (AI) uma kwenziwa kahle, kodwa umhlaba wezobuchwepheshe uhlukene ngokuthi "indlela efanele" ingaba yini.
I-AI, ikakhulukazi amamodeli ezilimi ezinkulu (ama-LLM) ngemuva kwe-AI ekhiqizayo (i-gen AI), ayikwazi ukubukwa ngendlela efanayo nohlelo lomthombo ovulekile. Ngokungafani nesofthiwe, amamodeli e-AI aqukethe ngokuyinhloko amamodeli epharamitha ezinombolo anquma ukuthi imodeli icubungula kanjani okokufaka, kanye nokuxhumana ekwenzayo phakathi kwamaphuzu edatha ahlukahlukene. Amapharamitha amamodeli aqeqeshwe angumphumela wenqubo ende ehilela inani elikhulu ledatha yokuqeqesha elungiswa ngokucophelela, ixutshwe, futhi icutshungulwe.
Nakuba amapharamitha emodeli engeyona isofthiwe, kwezinye izindlela anomsebenzi ofana nekhodi. Kulula ukuqhathanisa idatha nekhodi yomthombo yemodeli, noma into esondele kakhulu kuyo. Emthonjeni ovulekile, ikhodi yomthombo ivame ukuchazwa njengendlela "ekhethwayo" yokwenza izinguquko kwisofthiwe. Idatha yokuqeqesha yodwa ayihambisani nalo msebenzi, ngenxa yobukhulu bayo obuhlukahlukene kanye nenqubo eyinkimbinkimbi yangaphambi kokuqeqeshwa eholela ekuxhumaneni okubuthakathaka nokungaqondile okukhona noma iyiphi idatha esetshenziswa ekuqeqeshweni namapharamitha aqeqeshiwe kanye nokuziphatha okubangelwa yimodeli.
Iningi lokuthuthukiswa kanye nokuthuthukiswa kwamamodeli e-AI okwenzeka emphakathini okwamanje akubandakanyi ukufinyelela noma ukushintsha idatha yokuqeqeshwa yokuqala. Esikhundleni salokho, kuvela ekuguqulweni kwamapharamitha emodeli noma inqubo noma ukulungiswa okungase kusebenze nasekulungiseni ukusebenza kwemodeli. Inkululeko yokwenza lokhu kuthuthukiswa kwamamodeli idinga ukuthi amapharamitha akhishwe nazo zonke izimvume abasebenzisi abazitholayo ngaphansi kwamalayisense omthombo ovulekile.
Umbono we-Red Hat we-AI yomthombo ovulekile.
I-Red Hat ikholelwa ukuthi isisekelo se-AI yomthombo ovulekile sisezimisweni zemodeli ezinelayisensi yomthombo ovulekile ezihlanganiswe nezingxenye zesofthiwe yomthombo ovulekile . Lesi yindawo yokuqala ye-AI yomthombo ovulekile, kodwa akuyona indawo yokugcina yefilosofi. I-Red Hat ikhuthaza umphakathi womthombo ovulekile, iziphathimandla zomthetho, kanye nemboni ukuthi baqhubeke nokulwela ukucaca okukhulu kanye nokuhambisana nezimiso zokuthuthukiswa komthombo ovulekile lapho beqeqesha futhi belungisa amamodeli e-AI.
Lona umbono we-Red Hat njengenkampani ehlanganisa uhlelo lwesofthiwe yomthombo ovulekile futhi engakwazi ukusebenzisana ne-AI yomthombo ovulekile. Akuyona imizamo yokuthola incazelo esemthethweni, njengaleyo i- Open Source Initiative (OSI) eyithuthukisayo nge- Open Source AI Definition (OSAID). Lona umbono wenkampani mayelana nendlela yokwenza i-AI yomthombo ovulekile ibe nokwenzeka futhi ifinyeleleke emiphakathini, ezinhlanganweni, kanye nabathengisi abaningi ngangokunokwenzeka.
Lo mbono usetshenziswa ngokusebenza nemiphakathi yomthombo ovulekile, okugqanyiswe yiphrojekthi ye -InstructLab , eholwa yi-Red Hat, kanye nomzamo we-IBM Research on the Granite yomndeni wamamodeli omthombo ovulekile anelayisensi . I-InstructLab inciphisa kakhulu izithiyo kososayensi abangewona awedatha zokufaka isandla kumamodeli e-AI. Nge-InstructLab, ochwepheshe besizinda abavela kuzo zonke izinhlaka bangangeza amakhono nolwazi lwabo, kokubili ukusetshenziswa kwangaphakathi kanye nokusiza ukudala imodeli ye-AI yomthombo ovulekile eyabiwe futhi efinyeleleka kabanzi yemiphakathi engenhla.
Umndeni wamamodeli we-Granite 3.0 ubhekana nohlu olubanzi lwezimo zokusetshenziswa kwe-AI, kusukela ekwakhiweni kwekhodi kuya ekucutshungulweni kolimi lwemvelo kuya ekutholeni ulwazi oluvela kumasethi amakhulu edatha, konke ngaphansi kwelayisensi yomthombo ovulekile evunyelwe. Sisize i-IBM Research ukuletha umndeni wamamodeli wekhodi we-Granite ezweni lomthombo ovulekile futhi siqhubeke nokusekela umndeni wamamodeli, kokubili ngokombono womthombo ovulekile nanjengengxenye yokunikezwa kwethu kwe-Red Hat AI.
Imiphumela yezimemezelo zakamuva ze-DeepSeek ikhombisa ukuthi ukusungula izinto ezintsha okuvulelekile kungathinta kanjani i-AI, kokubili ezingeni lemodeli nangale kwalokho. Ngokusobala, kukhona ukukhathazeka ngendlela yesikhulumi saseShayina, ikakhulukazi ukuthi ilayisensi yemodeli ayichazi ukuthi yakhiqizwa kanjani, okuqinisa isidingo sokucaca. Lokho kusho ukuthi, ukuphazamiseka okukhulunywe ngakho kuqinisa umbono we-Red Hat ngekusasa le-AI: ikusasa elivulekile eligxile kumamodeli amancane, alungiselelwe, navulekile angenziwa ngokwezifiso ukuze asetshenziswe idatha yebhizinisi elithile kunoma iyiphi indawo ngaphakathi kwefu elihlanganisiwe.
Ukwandisa amamodeli e-AI ngale komthombo ovulekile.
Umsebenzi we-Red Hat esikhaleni se-AI somthombo ovulekile udlulela ngale kwe-InstructLab kanye nomndeni wamamodeli we-Granite, udlulela kumathuluzi namapulatifomu adingekayo ukuze kusetshenziswe futhi kusetshenziswe i-AI ngendlela ephumelelayo. Le nkampani isisebenze kakhulu ekukhuthazeni amaphrojekthi obuchwepheshe kanye nemiphakathi, njengokuthi (kodwa kungagcini lapho):
● I-RamaLama , iphrojekthi yomthombo ovulekile ehlose ukwenza lula ukuphathwa kwendawo kanye nokusetshenziswa kwamamodeli e-AI;
● I-TrustyAI , ithuluzi lomthombo ovulekile lokwakha imisebenzi ye-AI enesibopho kakhudlwana;
● I-Climatik , iphrojekthi egxile ekusizeni ukwenza i-AI ihlale isikhathi eside uma kuziwa ekusetshenzisweni kwamandla;
● I-Podman AI Lab , ithuluzi lonjiniyela eligxile ekwenzeni kube lula ukuhlola ngama-LLM omthombo ovulekile;
Isimemezelo sakamuva mayelana ne-Neural Magic sandisa umbono wenkampani we-AI, okwenza kube nokwenzeka ukuthi izinhlangano zivumelanise amamodeli e-AI amancane, alungiselelwe kahle, kufaka phakathi izinhlelo zemithombo evulekile ezinelayisensi, nedatha yazo, noma ngabe zihlala kuphi efwini elihlanganisiwe. Izinhlangano ze-IT zingasebenzisa -vLLM ukuze ziqhube izinqumo kanye nokukhiqiza kusuka kula mamodeli, okusiza ekwakheni i-AI stack esekelwe kubuchwepheshe obusobala nobusekelwayo.
Kuleyo nkampani, i-AI yomthombo ovulekile iphila futhi iphefumulela ifu elihlanganisiwe. Ifu elihlanganisiwe linikeza ukuguquguquka okudingekayo ukukhetha indawo engcono kakhulu yomsebenzi ngamunye we-AI, okwenza ngcono ukusebenza, izindleko, isikali, kanye nezidingo zokuphepha. Amapulatifomu, imigomo, kanye nenhlangano ye-Red Hat isekela le mizamo, kanye nabalingani bemboni, amakhasimende, kanye nomphakathi womthombo ovulekile, njengoba umthombo ovulekile kubuhlakani bokwenziwa uqhutshelwa phambili.
Kunethuba elikhulu lokukhulisa lokhu kubambisana okuvulekile esikhaleni se-AI. I-Red Hat ibona ikusasa elihlanganisa umsebenzi osobala kumamodeli, kanye nokuqeqeshwa kwawo. Kungakhathaliseki ukuthi kuyisonto elizayo noma ngenyanga ezayo (noma ngisho nangaphambi kwalokho, uma kubhekwa ukuthuthuka okusheshayo kwe-AI), inkampani kanye nomphakathi ovulekile wonkana bazoqhubeka nokusekela nokwamukela imizamo yokwenza intando yeningi nokuvula umhlaba we-AI.
