La Tsvagiridzo yeApple yekudzidzisa zvemichina Yave imwe yenzvimbo dzakakosha dzekambani, kunyangwe kwemakore yaionekwa sekambani yakavanzika kupfuura dzimwe hofori dzeAI. Zvisinei, nhasi uno, Apple inobatanidza simba guru reinjiniya, hardware yayo, uye kutarisa kwakajeka pakuchengetedzwa kwevashandisi kuti vagadzire mamodheru epamusoro anoshanda mugore uye zvakananga pamudziyo.
Kuseri kwemabasa atinoshandisa tisingafungi—kubva pane Kuziva inzwi reSiri Kubva pakuderedza ruzha rwemafoni kusvika pakushandisa mazano ekunyora nemaoko, kune injini yekudzidzisa michina ine hunyanzvi iri kushanda. Apple iri kutungamira kufambira mberi mukuvandudza ma network makuru e neural, ma generative models, natural language processing, uye open frameworks seMLX, nechinangwa chimwe chete: kuti AI ive nesimba, inoshanda zvakanaka, uye inoremekedza data revanhu.
Apple iri kutarisira kudzidza kwemuchina uye AI
Artificial intelligence uye kudzidza kwemuchina ndiwo mbiru mbiri dzakakosha mune ramangwana reApple. Makore mashoma apfuura, kambani iyi haina kutaura zvakawanda nezvekuvandudzwa kwayo mukati, asi maitiro iwayo achinja zvakanyanya. Nhasi, havangoshandisi ruzivo rwemuchina chete kuzvinhu zvavo zvese, asi vanobvumirawo mainjiniya avo kuburitsa basa ravo uye kutaurirana nevesainzi.
Kuchinja uku kwakaitika nekutanga kwe Apple Machine Kudzidza ChinyorwaIri iblog repamutemo apo mainjiniya vanotsanangura, neruzivo rwakadzama rwehunyanzvi, mashandisiro avanoita kudzidza kwemuchina kugadzira zvigadzirwa zvinoshandiswa nemamiriyoni evanhu. Chinangwa hachisi chekungotsanangura "zvavakaita" chete, asiwo "kuti" vakazviita sei, nemaitiro, kuyedza, uye mhedzisiro yehuwandu.
Mumamiriro ezvinhu aya, Apple yabva pakuonekwa sekambani yakavharwa muAI kusvika pakuva kambani yakavhurika, kuburitsa tsvakurudzo, kutora chikamu mumisangano, uye kuburitsa maturusi ekuvandudza. Shanduko iyi yakakosha kune kukwezva nekuchengetedza tarenda rekutsvagisanemifananidzo yakaita se John giannandrea, nguva apo vakwikwidzi vakaita seGoogle, OpenAI kana Meta vaive nemukana nekuti vaive netsika yakavhurika.
Panguva imwe chete, kambani iyi inochengetedza maitiro ayo akasiyana: zvese zvainoita mukudzidza kwemuchina zvakabatana zvakanyanya nebhizinesi rayo guru. hardware, operating system uye chigadzirwa dhizainiIzvi zvinoonekwa mumachipisi avo eApple Silicon uye munzira yavanosanganisa mamodheru zvakananga mu iOS, macOS, kana watchOS, pasina kumanikidza vagadziri kuti vagadzirise matambudziko asina basa.
Kusanganiswa kwekutsvaga kwemhando yepamusoro, nzira inoshanda, uye kunyanya kuda... ruzivo rwemushandisi rwakatarisana nekuchengetedzwa kwezvakavanzika Izvi ndizvo zvinosiyanisa Apple mukati meAI ecosystem, kunyangwe zvichienzaniswa nemamodeli makuru egore akadai seChatGPT kana mamwe maLLM makuru.
Manetwork makuru etsinga dzinodzokororwa uye kufambira mberi kweApple
Imwe yenzvimbo idzo Apple yakaita danho guru iri mu recurrent neural network (RNN)Kare, maRNN ainyanya kukwezva mabasa akatevedzana akadai semashoko kana odhiyo nekuti anoshandisa ndangariro shoma uye kuverenga kupfuura magadzirirwo akavakirwa pakucherechedza (akadai seTransformers). Zvisinei, maitiro avo akatevedzana akaita kuti zvive zvakaoma zvikuru kukwidza mamodheru aya kusvika kumabhiriyoni emaparameter.
Vaongorori veApple vakagadzira nzira inoita kuti kudzidzisa maRNN makuru kunoshanda zvakanyanyaIzvi zvinobvumira, kekutanga, dhizaini iyi kuti ishandiswe pazviyero zvaimbochengeterwa mamodheru akavakirwa pakutarisa chete. Nekugadzirisa mashandisirwo ehurongwa uye kuita kuti kuverenga kuve kwakafanana, vakaderedza zvakanyanya mari yekudzidzira uye vakaita kuti ishande pamidziyo iripo.
Izvi zvinonyanya kukosha kuApple nekuti maRNN akagadzirwa zvakanaka anobvumira fungidziro dzinokurumidza zvikuru uye dzisina ndangariro yakawandaIzvi zvakakosha pakushandisa mamodheru zvakananga pamidziyo yakaita se iPhone, iPad, kana Mac pasina kupedza bhatiri kana kuwedzera CPU zvakanyanya. Panzvimbo pekushandisa mamodheru makuru akavakirwa pamakore, Apple inotarisa pakugadzira marongerwo akamanikana uye anoshanda zvakanaka anogona kushanda munharaunda.
Rudzi urwu rwekutsvaga runoenderana zvakakwana nehurongwa hwavo hweML AI huripo pamudziyo, uko kuri pakati pe simba, kushandiswa uye kunonoka Zvakakosha. Modhi yakarurama zvikuru asi inoda zvinhu zvakawanda zvekushandisa pakombiyuta hainyanyi kunakidza kana isingashande zvakanaka pamamiriyoni emidziyo chaiyo.
Uyezve, kufambira mberi uku muRNN kunogona kusanganiswa nedzimwe nzira, dzakadai semamodeli ekugadzira kana ehybrid anoshandisa zvikamu zvinodzokororwa pamabasa akati uye kutarisirwa kwevamwe, zvichivhura mukana wekuti magadzirirwo anochinjika uye akagadzirwa nenzira yakajairika zvichienderana nekushandiswa.
Chinyorwa cheApple Machine Learning uye kuvhurwa kwekuferefeta
Kwenguva yakareba, Apple yaizivikanwa nekunyanya kwayo kuchengeta zvakavanzika uye kudzivirira mainjiniya ayo kuburitsa zvinyorwa zvesainzi. Mutemo uyu, sekureva kwenyanzvi dzakawanda, wakaita kuti kambani... vaizorasikirwa nemukurumbira kana zvichienzaniswa nemamwe makambani eAI,kukonzera a AI brain drainuko vaongorori vaigona kuburitsa, kupinda misangano, uye kuvaka mukurumbira wavo mune zvedzidzo.
Mamiriro ezvinhu aya akachinja zvakanyanya apo Apple yakazivisa, kuburikidza nemukuru wayo wezvekutsvagisa Russ Salakhutdinov, kuti yaizove yakavhurika zvakanyanya panyaya yekuburitsa mabhuku. Sechikamu chekuchinja ikoko, Apple Machine Kudzidza Chinyorwa, nzvimbo yepamutemo uko mainjiniya vanogovana mhedzisiro chaiyo kubva kumapurojekiti anozopedzisira ava zvigadzirwa zvevatengi.
Chinyorwa chekutanga cheblog chakanangana nekuvandudza kuoneka kwemifananidzo yakagadzirwaDambudziko raive pachena: kudzidzisa ma neural network nemifananidzo chaiyo, yakanyorwa nemaoko kunodhura uye kunononoka, nepo mifananidzo yekugadzira (yakagadzirwa nemaoko) yakachipa zvikuru kugadzira asi haiwanzotaridzika sechokwadi, izvo zvinoderedza mashandiro emamodheru.
Chikwata cheApple chakatsanangura nzira yeku "polisha" mifananidzo iyi yakagadzirwa kuti iite seyechokwadi, ukuwo ichichengetedza zvinhu zvakakosha zvinodiwa ne neural network. Neizvi, vakakwanisa kuita izvi. kuwedzera kururama kwemamodheru pasina kuvimba zvakanyanya nedata rakanyorerwa nemaoko, izvo zvinokurumidzisa zvikuru kugadzirwa kwemasisitimu ekuona emakombiyuta, semuenzaniso, ekuziva mifananidzo zviri nani mu iOS.
Kupfuura kuburitswa kwekutanga ikoko, Apple inoshandisa Journal senzvimbo yekutsanangura kufambira mberi mukuona, mutauro, kuziva kutaura, nedzimwe nzvimbo. Vanokokawo... vaongorori, vadzidzi nevagadziri kutumira mibvunzo nemhinduro kuburikidza neemail, kusimbisa hukama nenharaunda iyo makore mashoma apfuura yairatidzika kunge isingafungidzike.
Kuchinja uku kwemukati mune zvepfungwa hakungobatsiri vesainzi chete; asi inzira yekuti Apple ivandudze mabasa ayo, akadai semazano eApple Music, kugona kwaSiri kufanotaura, kana kugadzirisa mapikicha epamusoro pa iPhone, ayo anovimba zvakanyanya nemamodheru ekudzidza emuchina ari kuwedzera kuomarara.
Zvikwata zvekutsvagisa zvekudzidza kwemichina mukati meApple
Tsvagiridzo yekudzidza kwemichina paApple haisi chinhu chimwe chete, asi yakarongwa muzvikamu mapoka akasarudzika anofukidza nzvimbo dzakasiyana yeAI inoshandiswa. Pakati pezvinonyanya kuzivikanwa pane zvivakwa zveML, kudzidza kwakadzama uye kusimbisa, uye mapoka ekugadzira mutauro wechisikigo uye matekinoroji ekutaura.
Chikwata che zvivakwa zvekudzidza zvemichina Chikwata ichi chine basa rekuvaka hwaro hwehunyanzvi hwakavakirwa pazvigadzirwa zvitsva zvekambani. Vanoshanda nemapuratifomu makuru emakombiyuta, maturusi ekuongorora, uye masisitimu makuru ekuchengetera zvinhu, zvichivagonesa kudzidzisa nekushandisa mamodheru makuru zvinobudirira.
Boka iri rinoshanda pamwe chete nemamwe madhipatimendi eApple kuti vagadzirise mamiriro ezvinhu. hardware, software, uye maalgorithms yakagadzirirwa zvinodiwa zvechigadzirwa chimwe nechimwe. Inzvimbo inotsvakwa ma profiles mu data science, backer-end engineering, mapuratifomu nemasystem, pakati pezvimwe, uye uko chinhu chikuru chiri kukwanisa kupa AI yakasimba kumamiriyoni emidziyo.
Kune rumwe rutivi, chikwata che kudzidza zvakadzama uye kudzidza kusimbisa Inobatanidza vaongorori nemainjiniya vane hunyanzvi mumhando dzakasiyana-siyana dzedzidzo: kudzidza kunotarisirwa uye kusingatarisirwi, mamodheru ekugadzira, kudzidza kwenguva pfupi, kudzidza kwemhando dzakasiyana-siyana, kudzidza kwakadzika kwekusimbisa, kudzidza kwekusimbisa kwakapesana, dzidziso yesarudzo, uye dzidziso yemutambo.
Boka iri rinosundira miganhu ye hungwaru hwekugadzira hwakakuraKutsvaga mhinduro kumatambudziko chaiwo anoda mamodheru akasimba uye anoshanda. Chikamu chino chinoongorora zvese kubva pakuita kuti mumiriri adzidze kuita sarudzo dzakanakisa zvichibva pakushanda kwake nenzvimbo, kusvika pakubatanidza zvinyorwa, mifananidzo, uye odhiyo mumasisitimu ane modal akawanda zvechokwadi.
Pakupedzisira, chikwata che kugadzirisa mitauro yechisikigo uye matekinoroji ekutaura Inobatanidza masayendisiti anobva kumatunhu akasiyana-siyana, ese akatarisana nekunzwisisa nekugadzira mutauro wevanhu. Vanoshanda pamabasa akadai sekushandura muchina, kuziva mafambiro emazwi, kutsvaga mhinduro, kupatsanura misoro, uye kuziva kutaura, nemashandisirwo akajeka muSiri, kuraira, kushandura pachishandiso, uye vabatsiri vekutaura.
Mapoka aya anowanzo shanda neruzivo rwakawanda uye anoshandisa matekiniki epamusoro ekudzidza zvakadzama kugadzirisa matambudziko ari mu mitauro yakawanda uye matauriroVari kutsvaga ma profiles mu natural language engineering, language modeling, text-to-speech software, speech environment engineering, data science, uye pure research, zvichiratidza hupamhi hwebasa riri muApple.
Kudzidza Kwemuchina Pamudziyo uye Uchenjeri hweApple: AI pamudziyo
Chimwe chezvinhu zvinosiyanisa Apple nezvimwe vatambi vari muchikamu ichi ndechekuti inonyanya kufarira Apple. AI ichashanda zvakananga pamudziyo.pasina chikonzero chekutumira data rese ku cloud. Chikwata cheOn-Device Machine Learning chinopa maturusi kune vanogadzira mapurogiramu nevanofarira kubatanidza Apple Intelligence ne machine learning mumapurogiramu nemapurojekiti emunhu.
Muchirevo ichi, pfungwa dzakadai se ungwaru hwepuratifomuuko ML neAI zvisiri zvinhu zvekuwedzera chete, asi zviri chikamu chepakati pemasisitimu ekushanda. Nekuda kwetekinoroji idzi, zviitiko zvisina matambudziko zvakaita sekusimbiswa kwechokwadi, kuziva manyorerwo, uye kuderedza ruzha rwekufona zvinogoneswa, zvese zvichishanda zvakajeka kumushandisi.
Mumakore achangopfuura, Apple yakabatanidzawo njere dzinogadzira zvinhu dziri pakati pehurongwa hwadzo ine zvinhu zvakaita seWriting Tools, Genmoji, uye Image Playground. Zvishandiso izvi zvakagadzirirwa kuti zvibatanidzwe zviri nyore mumapurogiramu aripo, zvichibvumira vagadziri kuti vawedzere ma interfaces avo uye vawedzere zvinhu zvakangwara pasina kufanira kugadzira mamodheru kubva pakutanga.
Nechinangwa ichocho, Apple inopa zvakasiyana-siyana zve maAPI anoshandisa michina yekudzidzaMaAPI aya anopa mukana wekushandisa mamodheru nehunyanzvi hwesystem iyi. Mienzaniso inosanganisira ImageCreator yekugadzira mifananidzo, uye mamwe maAPI anogonesa mazano emhinduro akangwara, kupatsanura zviri mukati, kana kuongorora magwaro zvakananga pamudziyo.
Nekusvika kwechimiro cheFoundation Models mu iOS 26Maitiro aya anowedzera kurerutswa. Chimiro ichi chinoita kuti munhu akwanise kuwana modhi yemutauro wepamudziyo yakagadziriswa zvakanyanya uye nyanzvi mumabasa ezuva nezuva akadai sekupfupisa, kubvisa ruzivo, kuisa muzvikamu kana kugadzira zvinyorwa, zvese izvi uku uchichengetedza zvakavanzika zvemushandisi nekushanda usiri paindaneti zvachose.
MaAPI ekudzidza kwemuchina uye maframeworks muApple ecosystem
Kuti ive nechokwadi chekuti kufambira mberi uku kwese hakungogumiri mumarabhoritari chete, Apple inochengetedza uye inovandudza seti yakasimba ye MaAPI ekudzidza kwemuchina MaAPI aya anowanikwa kune vanogadzira mapurogiramu uye anofukidza zvese kubva pakuona kwekombiyuta nekuongorora ruzha kusvika pakugadzirisa nekushandura mitauro yechisikigo.
Pakati pezvivakwa zvikuru pane kuona, yakanangana nekuongorora mifananidzo nemavhidhiyo, iyo inosanganisira mabasa akadai sekuzivikanwa kwemagwaro, kuona zviso, kupatsanura zvinhu kana, pakati pezvinhu zvitsva zvitsva, kuona mavanga ari pane lenzi yekamera kuti zvinyatso gadzirisa kutora mifananidzo.
Mundima yemitauro, marongero akadai se Natural Languageiyo inobvumira kushandura masaini, kuona mutauro, kuongorora manzwiro, kupatsanura magwaro, uye kubvisa zvinhu; uye Translation, yakatarisana nekushandura pakati pemitauro yakawanda, ichinyanya kukoshesa kushandurwa kwemitauro yemuno kana zvichibvira kuchengetedza zvakavanzika.
Chikamu cheruzha nezwi chakagadzirwa zvakatenderedza Kuongorora Ruzha, iyo inoziva maitiro eruzha munharaunda, uye Kutaura, ine basa rekuziva uye kunyora manzwi. Chimwe chinhu chitsva chakakosha mubasa iri iSpeechAnalyzer API, iyo inogonesa kugadzirisa manzwi nemagwaro nekukurumidza uye nyore nyore, inonyanya kubatsira pakurekodha manzwi marefu kana kurekodha nemamaikorofoni ari kure.
Uyezve, vagadziri vanogona kugadzirisa mamodheru aya vachishandisa app ne framework. GadziraMLIzvi zvinoita kuti kudzidziswa kwemamodeli kuve nyore uchishandisa data riripo pasina kuda zvivakwa zvakaoma. Izvi zvinobvumira mapurojekiti madiki kusvika pakati nepakati kushandisa matekiniki ekudzidza kwemuchina pasina mari yakawanda yekutanga.
Core ML, Apple Silicon, uye mashandiro anoshanda emamodheru
Chinhu chikuru mukuvandudzwa kweAI muApple ndeichi Core MLCore ML ndiyo framework inoita kuti zvive nyore kubatanidzwa kwema machine learning models muma app e iPhone, iPad, Mac, Apple Watch, kana Apple TV. Core ML yakagadzirwa kuitira kuti vagadziri vagone kutarisa pane logic yeapplication yavo, ukuwo framework ichibata nekugadzirisa uye kushandisa modhi.
Mamodheru anoenderana neCore ML anogona kudhawunirodhwa atoshandurwa kubva developer.apple.com kana kubva pawebhusaiti yepamutemo yeHugging Face yeApple, kana kushandura kubva kune mamwe mafomati uchishandisa CoreML Tools. Seti iyi yezvishandiso inogadzirisa modhi yekushandisa pachishandiso, ichideredza saizi yacho uye ichivandudza mashandiro nehunyanzvi hwe otomatiki uye hwemaoko.
Kana yagadzirwa, modhi iyi inobatanidzwa zviri nyore muXcode. Inogona kuitwa zvakananga kubva panzvimbo yekuvandudza. tarisa magadzirirwo acho, ongorora mashandiro edzidziso pamidziyo yakasiyana uye gadzira Swift interfaces dzakachengeteka kune mhando dzemakombiyuta, zvichiita kuti zvive nyore kushandisa pasina zvikanganiso zvekunyora kana matambudziko ekubatana.
Panguva yekumhanya, Core ML inoshandisa nekuchenjera CPU, GPU uye Neural Injini Machipisi eApple Silicon anoshandiswa, achisarudza musanganiswa unoshanda zvakanyanya zvichienderana nebasa remota. Izvi zvinobvumira mamodheru akaomarara kuti ashande zvakanaka kunyangwe pamidziyo inotakurika, pasina kupedza bhatiri.
Kana paine kudiwa kwekutonga kwakanyanya, Apple inopa mukana wekubatanidza mamodheru eCore ML nemafuremu edanho repasi akadai se MPSGraphMetal Compute kana kuti Accelerate's BNNS Graph API. Pakati pezvavanogona kuita munguva pfupi yapfuura, BNNSGraphBuilder inoonekwa seyakakosha, yakagadzirirwa mabasa eML akavakirwa paCPU, uye yakakodzera kumapurogiramu anoda nguva shoma.
MLX uye kukurudzira kutsvaga paApple Silicon
Kumhanya kuri kuita tsvakurudzo yemichina yekudzidza kunoda maturusi ane simba uye anochinjika. Kuti zvizadzise chishuwo ichi, Apple yakagadzira MLX, hurongwa hwemakombiyuta enhamba hwakavhurika uye ML hwakagadzirirwa kushandisa magadzirirwo eApple Silicon.
MLX inokutendera kuti uite kubva kugadzirisa mamodheru ekudzidzisa nekudzidza kwakapararira yemamodeli echizvarwa chinotevera zvakananga mumidziyo yeApple. Nekuda kwekuvakwa kwendangariro kwakabatana kweApple Silicon, CPU neGPU zvinogona kushanda pane buffer imwe chete panguva imwe chete, zvichiita kuti kodhi ive nyore uye zvichivandudza mashandiro.
Chimwe chezvinhu zvinokwezva zveMLX ndechekuti inogona kushandiswa mu Python, Swift, C++ nemimwe mitaurokuchinjira kuzvinodiwa nevagadziri vakasiyana uye mapurofayiri ekutsvagisa. Uyezve, zvinoita kuti zvikwanisike kumhanyisa mamodheru emitauro mikuru nekufona kwemutsetse mumwe chete wemirairo, izvo zvinobatsira zvikuru pakuedza nekukurumidza.
Kunyangwe Apple ichikurudzira MLX senzira yayo, inopawo rutsigiro rwakasimba kune nzira dzakakurumbira dzakadai se PyTorch kana JAX kuburikidza neMetal, iyo ine magirafu emhando yepamusoro API. Izvi zvinoreva kuti vaongorori vanoziva nezve ecosystems idzi vanogona kushandisa simba reApple Silicon vasina kunyora patsva makodhi avo ese.
Kuti nharaunda igare ichiziva nezvezvinoitika, ine zvinhu zviripo zvakaita se developer.apple.com uye marepositories eGitHub eAppleuko maraibhurari, mienzaniso, uye magwaro matsva pamusoro pezvinhu zvitsva zvekudzidza kwemuchina mukati meApple ecosystem zvinoburitswa.
Apple Machine Learning sepuratifomu yevanogadzira uye masayendisiti edata
Pasi peApple Machine Learning pane boka re matekinoroji, maturusi uye masevhisi Yakagadzirirwa vese vanogadzira uye masayendisiti edata. Chinangwa ndechekugonesa kugadzirwa, kudzidziswa, uye kutumirwa kwemamodeli eAI nekukurumidza, zviri nyore, uye zvakachengeteka, pasina kukanganisa kunaka kwemhedzisiro.
Puratifomu iyi inopa musanganiswa we mamodheru akatsanangurwa kare uye mamodheru anogadziriswaIzvi zvinobvumira vashandisi kusarudza pakati pemhinduro dzisiri dzenguva dzose kana kugadzirisa mamodheru zvichienderana nezvinodiwa chaizvo. Izvi zvinokanganisa mabasa akadai sekuzivikanwa kwemifananidzo, kugadziriswa kwemitauro yechisikigo, ongororo yekufanotaura, uye mamwe maapplication akawanda akajairika eML.
Chimwe chezvinhu zvinosimbisa Apple Machine Learning ndechekutarisa kwayo pa kuchengeteka uye kuvimbikaZvese zvakagadzirirwa kuve nechokwadi chekuti mamodheru aya anochengetedza mwero wepamusoro wekururama uye kushanda, ukuwo achichengetedza ruzivo rwevashandisi kuburikidza nekushandisa pamudziyo pese pazvinogoneka uye mitemo yakasimba yekuchengetedza zvakavanzika.
Iyo interface nezvishandiso zvine chekuita nayo zvakagadzirwa kuti zvive zviri nyore kunzwisisa uye zvakabatana zvakanaka nebasa rekushanda Izvi ndizvo zvinoitika mukuvandudzwa kweXcode nedzimwe nzvimbo dzeApple. Izvi zvinovandudza kugona kwemushandisi uye zvinoderedza kukakavadzana kana uchitora modhi kubva pachikamu cheprototype kuenda kuapp inoshandiswa nemamiriyoni evanhu.
Ingave mapurojekiti madiki emunhu kana zvigadzirwa zvakaoma pasi rose, Apple Machine Learning inopa zvivakwa zvakasimba uye maturusi anochinjika anokugonesa kushandura mazano kuita mhinduro dzeAI dzechokwadi, uchinyatsotarisa pane ruzivo rwemushandisi wekupedzisira.
Zvikatorwa pamwe chete, zvidimbu zvakasiyana-siyana zvakagadzirwa neApple—kubva pakufambira mberi mumabasa ekudzokorora neural network kusvika kuma framework akaita seCore ML, maAPI akaita seVision and Natural Language, Research Journal, uye MLX yeApple Silicon—zvinoratidza nzira yakabatana: kupa AI ine simba uye inoshanda zvakanaka iyo yakabatanidzwa zvakanyanya muhardware, pasina kukanganisa zvakavanzika kana ruzivo rwemushandisi rwunoratidza kambani yacho.
