Ungayithola kanjani i-NPU ku-Copilot+ PC: Umhlahlandlela Ophelele Wokuthola Okuningi Ku-AI

Uma uzibuza ukuthi ungayisebenzisa kanjani ngokugcwele i-NPU ye-Copilot+ PC yakho , ufike endaweni efanele. Kulo mhlahlandlela, ngizochaza ngokuningiliziwe, ngesiSpanishi (eSpain), ukuthi yini okudingeka uyisebenzise, ​​ukuthi ungayifinyelela kanjani kusuka ku -Windows 11 kanye nezinhlelo zakho zokusebenza , ukuthi yimaphi ama-API okufanele uwasebenzise, ​​ukuthi yimaphi amafomethi emodeli ahambisanayo, nokuthi ungalinganisa kanjani ukusebenza ukuqinisekisa ukuthi ukusheshisa kwe-AI kusebenza ngendlela efanele.

Ngaphandle kwezici zobuchwepheshe, uzobona okuhlangenwe nakho kwangempela i-Copilot+ ekunika amandla (ukukhiqiza, impilo yebhethri, ubumfihlo), ukuthi i-NPU ifanelana kanjani ne-CPU kanye ne-GPU, nokuthi kwenzekani uma idivayisi yakho ingafinyeleli ku-40 TOPS edingwa yi-Microsoft ukuze uthole ulwazi oluphelele lwe-Copilot+. Ngamafuphi, uzothola umbono osebenzayo we-ecosystem yonke ngaphandle kokuxinwa yi-jargon yobuchwepheshe engadingekile.

Iyini i-Copilot+ PC futhi kungani i-NPU ishintsha imithetho?

I-Copilot+ PC ichaza uhlobo olusha lwama -laptop nama-desktop e-Windows 11 aklanyelwe eduze kweyunithi yokucubungula imizwa esebenza kahle kakhulu (i-NPU). Lawa ma-NPU akhethekile emisebenzini yokufunda okujulile—ukuhumusha ngesikhathi sangempela, ukukhiqiza/ukuhlela izithombe, imiphumela yevidiyo ye-AI—futhi ayakwazi ukudlula ama-TOPS angu-40 (ama-trillion okusebenza ngomzuzwana), okuvumela ukwenziwa kwamamodeli endawo ane-latency ephansi kanye nokusetshenziswa kwamandla okuncane.

Isihluthulelo ukuthi i-NPU isebenza kanye ne-CPU kanye ne-GPU . I-Windows 11 inika umsebenzi ngamunye umthombo ofanele kakhulu ukuze ilinganisele isivinini kanye nokusebenza kahle: i-CPU ye-logic ejwayelekile, i-GPU yehluzo/i-ML ehambisanayo, kanye ne-NPU yesilinganiso se-AI esiqhubekayo esinokusebenza okuhle kakhulu nge-watt ngayinye. Lokhu kusatshalaliswa yikho okuvumela imisebenzi ye-AI ebushelelezi ngaphandle kokuqeda ibhethri lakho phakathi kwasekuseni.

Kuhlelo lwe-Copilot+, iMicrosoft iletha okuhlangenwe nakho kwe-Windows AI kanye nama-API ahlanganiswe ku-Windows AI Foundry, ngokusekelwa kwamamodeli alungiselelwe asebenza ku-NPU. Lawa makhono ahlanganiswa kancane kancane ezinguqulweni zesimanje ze-Windows 11 kanye ne-Windows app SDK, okunciphisa ukungezwani kwabathuthukisi kanye nabasebenzisi bokugcina.

Umehluko uma uqhathaniswa ne-PC yendabuko? Nakuba i-GPU enamandla ingasheshisa i-AI, i-NPU ihlelwe kahle ukuze iphathe buthule futhi ngempumelelo imithwalo yesikhathi eside yokuphetha, okuholela empilweni yebhethri engcono kanye nokuphendula okuqhubekayo emisebenzini yangemuva ( amagama aqoshiwe abukhoma , ukufiphaza okuhlakaniphile, ukucindezela umsindo, njll.).

Izidingo Zedivayisi Nokuhambisana

Okuhlangenwe nakho okugcwele kwe-Copilot+ kudinga idivayisi ehambisanayo . I-Microsoft ibeke isilinganiso samandla e-NPU angu-40 ukuze kuqinisekiswe ukusebenza kahle nokusebenza kahle. Lesi silinganiso sesivele sihlangatshezwa amadivayisi anesizukulwane sakamuva se-Arm SoCs futhi siqhubeka nokwengezwa kumapulatifomu e-Intel ne-AMD ku-Windows 11.

Uma usebenza endaweni yobungcweti, kunezinhlobo ze-Copilot+ zezinga lebhizinisi (isibonelo, amadivayisi e-Surface aqondiswe ku-IT) anokuphepha kwezinga lebhizinisi kanye nezinzuzo ezifanayo zekhompyutha yendawo ye-AI. Kungakhathaliseki ukuthi umenzi ungubani, into ebalulekile ukuthi idivayisi yakho ifaka i-NPU esebenza kahle kanye ne-firmware/abashayeli abasesikhathini ukuze iWindows ikwazi ukusebenzisa izindlela zokusheshisa ezifanele.

Kuthiwani uma i-PC yakho ingafinyeleli ku-40 TOPS? Usazokwazi ukusebenzisa izici ze-AI, kodwa hhayi konke okufakiwe kuphakheji ye-Copilot+ . I-Windows ingasebenzisa i-GPU noma i-CPU njengenye indlela, yize lokhu kuzoholela ekusetshenzisweni okuphezulu kwezinsizakusebenza kanye nokubambezeleka isikhathi eside. Ngamanye amazwi, usazoba ne-AI, kodwa hhayi isipiliyoni esilungiselelwe, esiphelele esihlukanisa i-Copilot+.

I-Snapdragon X Elite namanye amapulatifomu: ukuthi umthwalo womsebenzi uhlukaniswa kanjani

I-Snapdragon X Elite SoC eyenziwe yi-Qualcomm, esekelwe ku-Arm architecture , ifaka le filosofi: ihlanganisa i-NPU ehamba phambili ekilasini ekwazi ukucubungula idatha enkulu ngesikhathi esifanayo nokusebenza kahle kwamandla okukhulu kakhulu kune-CPU/GPU ye-AI. Empeleni, lokhu kusho impilo yebhethri ende kanye nokushisa okuncane ngesikhathi somsebenzi wansuku zonke we-AI.

I-Windows 11 ilawula ukuhlelwa phakathi kwe-CPU, i-GPU, ne-NPU . Uma uvula uhlelo lokusebenza olunezici ze-AI, uhlelo lunquma ukuthi luthumele imisebenzi ku-NPU (ekhethwayo), i-GPU, noma i-CPU, lushintsha ngokusekelwe ekutholakaleni nasekuzinzeni kwabahlinzeki bokusebenza. Lokhu kwenzeka kalula; umane uqaphele ukuthi konke kuhamba kahle futhi ibhethri lakho lihlala isikhathi eside.

Kwezinye izinhlangothi, i-Intel ne-AMD nazo ziyathuthuka ngama-NPU ahlanganisiwe kanye nabahlinzeki bokusebenza abahambisanayo (i-OpenVINO ku-Intel, ama-EP azinikele ku-AMD) i-Windows engayivumela nge-Windows ML. Umgomo ohlanganyelwe ukunikeza ukusheshisa kwemvelo nokugcina ukuhambisana nehadiwe yangaphambilini kanye nabashayeli noma nini lapho kungenzeka.

Izici ze-AI ezikhethekile ku-Copilot+ nama-API atholakalayo

Ama-PC e-Copilot+ afaka phakathi ulwazi lwe-AI olwakhelwe ku-Windows 11 futhi olutholakala nge-Windows AI Foundry/Windows Runtime APIs yamamodeli e-NPU alungiselelwe. Lokhu kuhlanganisa konke kusukela emiphumeleni yekholi yevidiyo (ukufometha, ukufiphaza, ukukhansela umsindo) kuya ekufinyezweni kwendawo, ukuhumusha, kanye namakhono okukhiqiza, konke okuklanyelwe ukusebenza ngokuzimela ngaphandle kwefu.

Kwabathuthukisi, indlela yokuphetha ekhethwayo yi-Windows Machine Learning (Windows ML) . I-Microsoft ithutha ukufinyelela okunconywayo kusuka ku-DirectML kuya ku-Windows ML ukuze kube lula ukuthunyelwa, iphathe ngokuzenzakalelayo abahlinzeki bokusebenza (ama-EP), futhi igcine i-ONNX Runtime njengenjini yokuphetha eyisisekelo ngaphandle kokudinga ukuthi ulwe nama-binary kanye nokuncika.

Ukufinyelela ku-NPU ku-Copilot+ PC yabasebenzisi nabathuthukisi

Njengomsebenzisi, ungahlola ukusetshenziswa kwakho kwe-NPU ngesikhathi sangempela kusuka ku -Task Manager . Vula ithebhu ethi Ukusebenza futhi uzobona igrafu ye-NPU eceleni kwe-CPU, i-GPU, imemori, idiski, kanye nenethiwekhi. Yenza kusebenze, isibonelo, imiphumela yesitudiyo se-webcam futhi uzobona umsebenzi we-NPU oyimfihlo noqhubekayo, ofanele izingcingo zevidiyo ngaphandle kokukhipha ibhethri.

Njengonjiniyela, i-NPU iyisisetshenziswa sehadiwe okufanele usiqondise usebenzisa ama-API afanele. Ama-NPU aklanyelwe imisebenzi yenethiwekhi yesimanje ye-neural (convolutions, activation, attention, njll.), futhi ukufinyelela kwawo ku-Windows okwamanje kudluliselwa nge-Windows ML nge-ONNX Runtime ngaphansi, okuqinisekisa indlela yokusheshisa engcono kakhulu etholakalayo emshinini ngamunye.

Ukufinyelela Uhlelo nge-Windows ML: ama-EPs, ama-ORT, kanye ne-Fallback

I-Windows ML yethula ukutholwa nokulethwa kwabahlinzeki bokusebenza abakhelwe ngaphakathi (ama-EP) . Akusekho sidingo sokupakisha ngesandla i-QNNexecutionProvider ye-Qualcomm, i-OpenVINO EP ye-Intel, noma ezinye: I-Windows iyazihlanganisa noma izisebenzela nge- Windows Update , okunciphisa usayizi wohlelo lokusebenza nezinkinga zokuncika.

Ngaphansi, i-ONNX Runtime (ORT) ihlala iyinjini yokuphetha evulekile esebenzisa amamodeli akho e-ONNX. I-Windows ML isusa ubunzima: ihlola ihadiwe etholakalayo, ikhethe i-EP efanele (i-QNN uma i-Qualcomm NPU itholakala; i-OpenVINO uma ikhona; i-GPU/CPU njengesipele), ilanda/ilayisha umhlinzeki, bese iqala ukuphetha. Uma i-EP ekhethwayo yehluleka noma ingatholakali, ibuyela emuva ngokuzenzakalelayo kwenye indlela ngaphandle kokuphula uhlelo lwakho lokusebenza.

Lokhu kusebenzisana kusekelwa umsebenzi oqondile we-Microsoft nabakhiqizi (Qualcomm, Intel, AMD, njll.) ukuqinisekisa ukuhambisana komshayeli we-backward kanye nokusekelwa kwe-silicon entsha (isb., i-Snapdragon X Elite, i-Intel Core Ultra ), okukuvumela ukuthi ugxile kokuhlangenwe nakho kunokuhlanganiswa kwezinga eliphansi.

Amamodeli ahambisanayo, ukulinganisa, nokuhlanganiswa ne-Olive

Amamodeli amaningi aqeqeshwa ngokunemba okuphezulu njenge-FP32 , kodwa ama-NPU amaningi asebenza kahle kakhulu ngamanani amancane, ngokuvamile i-INT8. Ngakho-ke, imodeli ivame ukuguqulwa noma ukulinganiswa ukuze isetshenziswe ku-NPU, iphindaphinde ukusebenza kanye nokusebenza kahle ngaphandle kokulahlekelwa ikhwalithi eningi kakhulu.

Uma ungasebenzisi imodeli elungiselelwe kusengaphambili, ungaletha eyakho (BYOM) bese uyiqhuba nge-Olive toolchain, ecindezela, ilungise, futhi ihlanganise i-ONNX Runtime nge-NPU acceleration. I-Olive yenza kube lula izinyathelo ezazidinga izikripthi kanye nokulungiswa kwe-per-EP, futhi isheshise isikhathi sokukhiqiza ngokulungiswa kokusebenza okuzenzakalelayo.

Ungakala kanjani ukusebenza kwamamodeli we-NPU ne-AI

Ukuze uqinisekise ukuthi ukuhlanganiswa kwakho kuyashesha, udinga ama-metrics kanye ne-traces . I-Windows inikeza isethi enamandla yamathuluzi aqopha umsebenzi we-NPU, alinganise izikhathi zokuphetha, futhi ahlukanise izithiyo ngomqhubi, iseshini, noma umhlinzeki wokwenza.

Phakathi kwamakhono abalulekile, uzokwazi ukuqopha umkhondo wesistimu ngesikhathi sokusetshenziswa kohlelo lwakho lokusebenza, ubone ngeso lengqondo ukusetshenziswa kwe-NPU kanye ne-call stack yayo, uhlobanise umsebenzi ku-CPU/GPU/NPU, uhlaziye izikhathi zokulayisha kanye nokuqalisa (ukulayisha imodeli kanye nokudalwa kweseshini ye-ORT), ubuyekeze amapharamitha wokucushwa kwe-EP, kanye nephrofayili yabaqhubi ngabanye ukuze baqonde umnikelo wabo esikhathini esiphelele.

Ngaphezu kwalokho, imicimbi ye-ONNX Runtime ku-Windows Performance Analyzer (eqala nge-ORT 1.17, ngokuthuthukiswa ku-1.18.1) ikuvumela ukuthi ubuke izikhathi zokulayisha imodeli, izilungiselelo ze-EP, izikhathi zokuphetha, izingxenye ezithile (njenge-QNN), kanye namaphrofayili opharetha. Kuyisifinyezo esinembile salokho okwenziwa imodeli yakho kungqimba ngayinye.

Amathuluzi anconyiwe okuxilongwa kanye nokwenza iphrofayela

• Umphathi Womsebenzi: Ukubuka konke okusheshayo, kwesikhathi sangempela sohlelo lwakho (CPU, inkumbulo, idiski, inethiwekhi, i-GPU, kanye ne-NPU manje), namaphesenti okusetshenziswa, inkumbulo eyabiwe, inguqulo yomshayeli, nokuningi. Ilungele ukuqinisekisa ukuthi umsebenzi wakho uvula i-NPU.
• I-Windows Performance Recorder (WPR): Manje ihlanganisa iphrofayela ye-'Neural Processing' erekhoda ukusebenzisana kwe-Microsoft Compute Driver Model (MCDM) ne-NPU. Lokhu kukuvumela ukuthi ubone ukuthi yiziphi izinqubo ezisebenzisa i-NPU nokuthi yiziphi izingcingo ezithumela umsebenzi. Lokhu kuyasiza ekuhlukaniseni ukuhlehla noma ukuqinisekiswa kwe-EP.
• I-Windows Performance Analyzer (WPA): Iguqula ukulandelelwa kwe-ETW kube amagrafu namashadi adlula isikhathi ukuze kuhlaziywe i-CPU, idiski, inethiwekhi, imicimbi ye-ORT, kanye neshadi eliqondene ne-NPU, konke ngesilinganiso sesikhathi esifanayo. Kuyithuluzi elimaphakathi lezigaba zokuhlobanisa (ukulanda ngaphambilini/okungemuva kokulanda) nokubuka isithombe esikhulu sokusebenza.
• I-GPUView: Ifunda imicimbi ye-kernel nevidiyo kumafayela we-.etl futhi izethule ngokubonakalayo. Isekela ukusebenza kwe-GPU ne-NPU kanye nokubonwa kwemicimbi ye-DirectX kumadivayisi e-MCDM njenge-NPU. Iwusizo kakhulu uma ipayipi lakho lixuba ihluzo ne-ML.
• Iphrofayili ye-Qualcomm Snapdragon (qprof): Isixazululo esibanzi sephrofayili esichaza nge-NPU sub-HW (umkhawulokudonsa, izinto zokubala), kanye ne-CPU/GPU/DSP. Uma usebenza ku-Snapdragon X Elite, inikeza amasiginali abalulekile wokulungisa kahle.

Ukusebenza Okuphezulu: Kusho ukuthini nokuthi ungakuthola kanjani ukusebenza kweqembu lakho

I-TOPS (ama-trillion okusebenza ngomzuzwana) ibala ukuthi zingaki imisebenzi iprosesa engayenza ngeyunithi yesikhathi ngefomethi yezinombolo ethile. I-Microsoft isebenzisa i-TOPS engu-40 njengesilinganiso sesitifiketi se-Copilot+, ikunikeza umbono wezinhlobo zamamodeli nemiphumela ongayisebenzisa kahle endaweni.

Ukuze uthole amakhono e-NPU yakho, qala ngokukhomba iprosesa yekhompyutha yakho ku-'Izilungiselelo > Uhlelo > Mayelana'. Bese, bheka isilinganiso se-TOPS esisemthethweni kuwebhusayithi yomkhiqizi. Ukuze uthole ukuqhathanisa okwengeziwe kobuchwepheshe, amathuluzi afana ne-Procyon AI Benchmark akuvumela ukuthi ulinganise futhi uqhathanise i-NPU yakho namanye ama-NPU, yize lawa ehloselwe kakhulu ochwepheshe.

Sebenzisa Isiphathi Somsebenzi ukuze ubone ukusebenza kwe-NPU

Ngale kwezilinganiso zokusebenza eziphezulu zethiyori, bona ngeso lengqondo ukusetshenziswa kwe-NPU kwangempela nge-Task Manager: vula 'Ukusebenza' bese uya esigabeni 'se-NPU'. Uzobona igrafu yomsebenzi kanye nezilinganiso ezihlobene. Nika amandla izici ezifana nemiphumela yesitudiyo se-webcam ukuqinisekisa ukuthi ukuhamba komsebenzi wakho ku-NPU hhayi i-CPU/GPU.

Ukuze uthole ukuxilongwa okujulile, hlanganisa i-WPR/WPA namaphrofayili afanele . Ukuhamba komsebenzi okuvamile kungaba: ukulanda amaphrofayili 'ort.wprp' kanye 'etw_provider.wprp', qala ukuthwebula kanye namaphrofayili 'e-NeuralProcessing' kanye ne-CPU, phinda udlale ikesi, misa ukuthwebula, bese uvula ifayela le-.etl ku-WPA.

Ngokwesibonelo, kusuka ku-console ongayisebenzisa: wpr -start ort.wprp -start etw_provider.wprp -start NeuralProcessing -start CPU, dlala isimo sakho bese uqeda ngaso wpr -stop onnx_NPU.etl -compress. Bese uvula ifayela ku-WPA bese uhlola okuthi 'Neural Processing: NPU Utilization' kanye 'ne-Generic Events ye-ONNX' ukuze wenze izithenjwa eziphambanayo, isikhathi, nemicu.

I-Copilot+ empilweni yansuku zonke: ukukhiqiza, impilo yebhethri, nobumfihlo

Ekusetshenzisweni kwansuku zonke, i-Copilot+ isebenza njengomsizi womuntu siqu ohlanganiswe ku-Windows: idweba ama-imeyili namadokhumenti, ifingqa imibhalo emide, ilungisa izilungiselelo, ithole amafayela, futhi yenze imisebenzi ngokuzenzakalela isebenzisa ulimi lwemvelo. Ukhiye we -Copilot kwamanye amakhompyutha (njenge-ASUS) unikeza ukufinyelela okusheshayo ngaphandle kokuvula amamenyu.

I-NPU iphinde ithuthukise ulwazi lwe-multimedia kanye nokubambisana : ukufiphaza nokuphinda kugxile ngemuva ngesikhathi sangempela, imibhalo engezansi ezenzakalelayo, ukucindezela umsindo okuhlakaniphile, kanye nokuthuthukiswa kwamavidiyo okwakudinga isofthiwe edinga izinsiza eziningi ngaphambilini. Konke lokhu kwenzeka endaweni, ngokushelela nangokuthula, ngenxa yokulayisha umsebenzi we-NPU ku-CPU/GPU.

Enye inzuzo ecacile impilo yebhethri: ukuphathwa kwamandla okuhlakaniphile kanye ne-NPU kukhulisa ukusetshenziswa kwebhethri . Amamodeli afana ne-ASUS Zenbook A14 abonise izibalo zokusetshenziswa ezinde ezikuvumela ukuthi usebenze usuku lonke nge-AI esebenza ngemuva ngaphandle kokudinga ukuxhumeka njalo. Uhlelo luyazi ukuthi kufanele lucime nini futhi kufanele lubambe nini.

Ngokuphathelene nobumfihlo nokuvikeleka, ukusebenzisa i-AI kudivayisi kunciphisa ukuthembela efwini . Izici ezifana nokuqashelwa kwezwi noma ukuhlaziywa kwedokhumenti yomuntu siqu zingahlala ku-PC yakho. Amadivayisi afana ne-ASUS Vivobook anezela ama-shutter e-webcam angokoqobo kanye nokungena ngemvume kweminwe ukuze kuvikelwe kangcono.

Ukuhambisana, amahlebezi, kanye necala le-Intel/AMD/Qualcomm

Ingxenye yengxoxo yamanje igxile emfunweni we-TOPS engu-40 ye-Copilot+ . Kuvele amahlebezi okuthi amanye ama-CPU edeskithophu (isb., i-Intel Core Ultra 'Arrow Lake-S Refresh' ezayo) azogcina ama-NPU acishe abe yi-13 TOPS, okuzowavimbela ekuhlangabezaneni nelebula le-Copilot+ ngaphandle kosizo lwangaphandle.

Lokho akusho ukuthi 'awukwazi ukusebenzisa i-AI': ideskithophu ene-GPU ezinikele ingalungisa ukuntuleka kwamandla futhi isebenzise imithwalo yemisebenzi ye-AI esindayo, idlule kalula ama-TOPS angu-300 ku-INT8 ngamakhadi afana ne-RTX 4070. Umehluko ukuthi i-Copilot+ ibeka phambili ukusebenza kahle kwe-NPU okuqhubekayo kokuhlangenwe nakho okuhlala kuvuliwe, okunamandla aphansi.

I-Qualcomm , ngakolunye uhlangothi, isivele inikeza ama-NPU ahambisana ne-Copilot+ kumapulatifomu e-Arm njenge-Snapdragon X Elite, kanti i-Microsoft isebenzisana ne-Intel ne-AMD ukuqinisekisa ukuthi i-Windows ML nama-EP (i-OpenVINO, njll.) asebenza kahle kokubili ngehadiwe yamanje neyesikhathi esizayo. I-ecosystem ishintsha ngokushesha.

Izinsiza eziwusizo nalapho ungafunda khona okwengeziwe

Uma ufuna isingeniso esicacile semiqondo eyinhloko njenge- CPU, i-GPU, i-NPU, ne-TOPS , ukukhiqiza nge-Copilot+, ukusebenza, kanye nokuphepha, sincoma ukuthi ubukeze izinto zokuqeqesha kanye namaphepha edatha omkhiqizi. Isibonelo, ungabheka lesi sithombe esivela ku-AMD: Landa i-PDF.

Kwabathuthukisi, funda i-Windows ML kanye ne-ONNX Runtime ukuze usebenzise amamodeli endaweni, usebenzisa i-NPU. Uma uletha imodeli yakho, sebenzisa i-Olive chain ukuze uthole ukulinganisa nokwenza ngcono. Futhi ungakhohlwa ukusebenzisa izinsimbi: Imicimbi ye-WPR/WPA, i-GPUView, kanye ne-ORT (kusukela kunguqulo 1.17 kuqhubeke) ingabangane bakho bokushintsha kusuka 'emisebenzini' kuya 'ezimpukaneni'.

Uma uvela ekubukeni kuqala kwe-Copilot ku-PC endala (njenge-Skylake), yebo, i-NPU ingasetshenziswa njenge-accelerator njenge-GPU, kodwa ingcono uma imithwalo iqhubeka . I-Copilot+ ingeza ungqimba lokusebenza kahle kanye nokuqhubeka okunikezwa yi-NPU futhi yandisa ulwazi lwendawo; kumishini engenayo i-NPU (40 TOPS), i-Windows izozama ukuphatha umthwalo isebenzisa i-CPU/GPU, yize impilo yebhethri kanye nokubambezeleka ngeke kufane.

Uma utshala imali kukhompyutha entsha, bheka ngale kwe-CPU/GPU bese unaka i-NPU kanye nezinkomba zayo zokusebenza eziphezulu (TOPs ). Uma usuvele unayo i-Copilot+ PC, vumela izici ze-AI, qapha i-NPU ku-Task Manager, futhi uma ungumthuthukisi, thuthela ku-Windows ML ene-ORT ukuqinisekisa ukuthi uhlelo lukhetha indlela engcono kakhulu (QNN/OpenVINO) futhi unamaphrofayili athembekile ku-WPR/WPA.