I-Apple, ngokuvamile engaziwa ngokuvuleleka kwayo, ikhiphe imodeli ye-AI ekhiqizayo ebizwa nge-OpenELM ngokusobala eyenza kahle kakhulu kunesethi yezinye izinhlobo zezilimi eziqeqeshwe kumasethi edatha yomphakathi.

Akukona kakhulu - uma kuqhathaniswa I-OLMo, eyaqala ngoFebhuwari, VulaELM inembe kakhulu ngamaphesenti angu-2.36 kuyilapho usebenzisa amathokheni okuqeqesha kusengaphambili angu-2x ambalwa. Kepha mhlawumbe kwanele ukukhumbuza abantu ukuthi i-Apple ayisanelisekile ukuba yimbali yodonga embonini ye-AI.

Isimangalo sika-Apple sokuvuleleka sivela esinqumweni sayo sokungakhiphi nje imodeli, kodwa nohlaka lwayo lokuqeqesha nokuhlola.

“Ukuphambuka ezenzweni zangaphambili ezinikeza kuphela izisindo eziyimodeli nekhodi yokukhomba, kanye nokuziqeqesha kusengaphambili kumasethi edatha ayimfihlo, ukukhishwa kwethu kufaka uhlaka oluphelele lokuqeqeshwa nokuhlolwa kwemodeli yolimi kumasethi edatha atholakala esidlangalaleni, okuhlanganisa amalogi okuqeqesha, izindawo zokuhlola eziningi, kanye nangaphambi kwesikhathi. -ukuqeqeshwa ukucushwa,” kuchaza abacwaningi be-Apple abayishumi nanye kokuhambisanayo iphepha lobuchwepheshe.

Futhi ngokwehlukana nomkhuba wezemfundo, amakheli e-imeyili ababhali awafakiwe ohlwini. Chaza ekuchazeni ukuvuleka kwe-Apple, okucishe kuqhathaniswe ne-OpenAI engavuli kakhulu.

Okuhambisana nalokhu ukukhishwa kwesoftware akuyona ilayisensi eyaziwayo yomthombo ovulekile. Ayikhawuli ngokungafanele, kodwa ikwenza kucace ukuthi i-Apple iyaligodla ilungelo lokufaka isimangalo selungelo lobunikazi uma noma yimuphi umsebenzi osuselwe ku-OpenELM uthathwa njengowephula amalungelo ayo.

I-OpenELM isebenzisa indlela ebizwa ngokuthi isikali esihlakaniphile ngokwengqimba ukuze yabe amapharamitha ngokuphumelelayo kumodeli yesiguquli. Ngakho esikhundleni sokuthi isendlalelo ngasinye sibe nesethi efanayo yamapharamitha, izendlalelo zesiguquli se-OpenELM zinokucushwa okuhlukile namapharamitha. Umphumela uba ngcono Ukunemba, eboniswa kumaphesenti ezibikezelo ezifanele ezivela kumodeli ekuhlolweni kwebhentshimakhi.

Sitshelwa ukuthi i-OpenELM yayiqeqeshwe ngaphambilini isebenzisa i I-RedPajama Idathasethi evela ku-GitHub, ithani lezincwadi, i-Wikipedia, okuthunyelwe kwe-StackExchange, amaphepha e-ArXiv, nokuningi, kanye eyayigcwele isethi esuka ku-Reddit, Wikibooks, Project Gutenberg, nokuningi. Imodeli ingasetshenziswa ngendlela ongayilindela: Uyinikeza ukwaziswa, futhi izama ukuyiphendula noma ukuyiqedela ngokuzenzakalelayo.

Isici esisodwa esiphawulekayo sokukhishwa ukuthi kuhambisana "nekhodi yokuguqula amamodeli abe ngumtapo wezincwadi we-MLX ukuze aqondiswe futhi ahlelwe kahle kumadivayisi we-Apple."

I-MLX wuhlaka olukhishwe ngonyaka odlule lokusebenzisa umshini wokufunda ku-Apple silicon. Amandla okusebenza endaweni kumadivayisi e-Apple, esikhundleni sokusebenzisa inethiwekhi, kufanele enze i-OpenELM ithandeke kakhulu konjiniyela.

"Ukukhishwa kwe-Apple's OpenELM kuphawula inqubekelaphambili ebalulekile yomphakathi we-AI, ehlinzeka ngokusebenza kahle, ekusetshenzisweni kwe-AI esetshenziswa kudivayisi efanelekile kuzinhlelo zokusebenza zeselula namadivayisi e-IoT anamandla alinganiselwe wekhompyutha," u-Shahar Chen, oyi-CEO kanye nomsunguli we-AI service biz Aquant, utshele. IRejista. "Lokhu kwenza kube lula ukwenza izinqumo zasendaweni ezibalulekile kuyo yonke into kusuka kuma-smartphones kuya kumadivayisi ahlakaniphile asekhaya, kwandise amandla e-AI kubuchwepheshe bansuku zonke."

I-Apple izimisele ukukhombisa ukufaneleka kwesakhiwo sayo se-chip esikhule ekhaya sokufunda ngomshini, esisekelwa ikakhulukazi ku-Hardware selokhu iCupertino yethula Injini yeNeural ngo-2017. Noma kunjalo i-OpenELM, nakuba ingase ithole amaphuzu aphezulu kumabhentshimakhi wokunemba, ifika ifushane ngokuya ngokusebenza.

"Naphezu kokunemba okuphezulu kwe-OpenELM kwesibalo esifanayo sepharamitha, sibona ukuthi ihamba kancane kune-OLMo," kuchaza iphepha, licaphuna izivivinyo ezenziwa kusetshenziswa i-Nvidia's CUDA ku-Linux kanye nenguqulo ye-MLX ye-OpenELM ku-Apple Silicon.

Isizathu sokubonisa okungaphansi kokunqobayo, ama-boffins akwa-Apple athi, “ukusetshenziswa kwawo okungenangqondo kwe RMSNorm,” inqubo yokwenza idatha ibe yejwayelekile ekufundeni komshini. Ngokuzayo, bahlela ukuhlola ukulungiselelwa okwengeziwe.

I-OpenELM itholakala kumamodeli acushwe ngaphambili nacushwe yiziyalo anezigidi ezingama-270, 450 million, 1.1 billion kanye namapharamitha ayizigidi eziyizinkulungwane ezintathu. Abayisebenzisayo bayaxwayiswa ukuthi bakhuthale ngaphambi kokuzama imodeli nganoma yini enengqondo.

"Ukukhishwa kwamamodeli we-OpenELM kuhlose ukunika amandla kanye nokucebisa umphakathi wocwaningo ovulekile ngokunikeza ukufinyelela kumamodeli wezilimi ezisezingeni eliphezulu," kusho leli phepha. "Eqeqeshwe ngamasethi edatha atholakala esidlangalaleni, lawa mamodeli enziwa atholakale ngaphandle kwanoma yiziphi iziqinisekiso zokuphepha." ®

• I-SEO Powered Content & PR Distribution. Khuliswa Namuhla.
• I-PlatoData.Network Vertical Generative Ai. Zinike Amandla. Finyelela Lapha.
• I-PlatoAiStream. I-Web3 Intelligence. Ulwazi Lukhulisiwe. Finyelela Lapha.
• I-PlatoESG. Ikhabhoni, I-CleanTech, Amandla, Environment, Ilanga, Ukuphathwa Kwemfucuza. Finyelela Lapha.
• I-PlatoHealth. I-Biotech kanye ne-Clinical Trials Intelligence. Finyelela Lapha.
• Source: https://go.theregister.com/feed/www.theregister.com/2024/04/24/apple_openelm_ai/