Intel AI Analytics Toolkit for Linux

Ulwazi Lomkhiqizo

I-AI Kit iyikhithi yamathuluzi ehlanganisa izindawo eziningi ze-conda zokufunda ngomshini namaphrojekthi okufunda ajulile. Kubandakanya izindawo ze-TensorFlow, PyTorch, kanye ne-Intel oneCCL Bindings. Ivumela abasebenzisi ukuthi balungiselele uhlelo lwabo ngokusetha okuguquguqukayo kwemvelo, besebenzisa i-Conda ukwengeza amaphakheji, ukufaka abashayeli bezithombe, nokukhubaza i-hangcheck. Ikhithi yamathuluzi ingasetshenziswa ku-Command Line Interface (CLI) futhi ingahlanganiswa kalula kumaphrojekthi akhona ngaphandle kokulungiswa okukhethekile.

Ukusetshenziswa komkhiqizo

1. Lungiselela isistimu yakho ngokusetha okuguquguqukayo kwendawo ngaphambi kokuqhubeka.
2. Ukuze usebenze ku-Command Line Interface (CLI), sebenzisa iskripthi se-setvars.sh ukuze ulungiselele amathuluzi ku-toolkits ye-oneAPI ngokusebenzisa okuguquguqukayo kwemvelo. Ungathola umbhalo we-setvars.sh kanye ngeseshini noma ngaso sonke isikhathi uma uvula iwindi letheminali entsha. Iskripthi se-setvars.sh singatholakala kufolda yempande yokufakwa kwakho kwe-oneAPI.
3. Yenza kusebenze izindawo ezihlukene ze-conda njengoba kudingeka ngomyalo othi “conda activate ”. I-AI Kit ihlanganisa izindawo ze-conda ze-TensorFlow (CPU), i-TensorFlow ene-Intel Extension ye-S.ample TensorFlow (GPU), i-PyTorch ene-Intel Extension ye-PyTorch (XPU), kanye ne-Intel oneCCL Bindings ye-PyTorch (CPU).
4. Hlola imvelo ngayinye ehlobene ne-Getting Started Sampixhunywe kuthebula elinikezwe kumanuwali yomsebenzisi ukuze uthole ulwazi olwengeziwe mayelana nendlela yokusebenzisa indawo ngayinye.

Imiyalo elandelayo ithatha ngokuthi ufake isofthiwe ye-Intel® oneAPI. Sicela ubheke ikhasi le-Intel AI Analytics Toolkit ukuze uthole izinketho zokufaka. Landela lezi zinyathelo ukuze wakhe futhi usebenze njengeampne-Intel® AI Analytics Toolkit (AI Kit):

1. Lungiselela isistimu yakho.
2. Yakha futhi Ugijime i-Sample.

QAPHELA: Ukufakwa okujwayelekile kwePython kuhambisana ngokugcwele ne-AI Kit, kodwa i-Intel® Distribution yePython* iyathandwa.
Azikho izinguquko ezikhethekile kumaphrojekthi akho akhona adingekayo ukuze uqale ukuwasebenzisa nale khithi yamathuluzi.

Izingxenye Zaleli Kithi Yamathuluzi

I-AI Kit ihlanganisa

• I-Intel® Optimization ye-PyTorch*: I-Intel® oneAPI Deep Neural Network Library (oneDNN) ifakiwe ku-PyTorch njengomtapo wezincwadi ozenzakalelayo we-math kernel wokufunda ngokujulile.
• I-Intel® Extension ye-PyTorch:I-Intel® Extension ye-PyTorch* inweba amakhono e-PyTorch* ngezici zakamuva kanye nokuthuthukiswa kokusebenza okwengeziwe kwezingxenyekazi zekhompuyutha ze-Intel.
• I-Intel® Optimization ye-TensorFlow*: Le nguqulo ihlanganisa ama-primitives kusuka ku-oneDNN ukuya ku-Runtime ye-TensorFlow ukuze usebenze ngokusheshisa.
• I-Intel® Extension ye-TensorFlow: I-Intel® Extension ye-TensorFlow* iyi-plugin ehlukile, esebenza ngokujulile yokufunda esekelwe kusixhumi esibonakalayo se-TensorFlow PluggableDevice. Le plugin yesandiso iletha amadivayisi e-Intel XPU (GPU, CPU, njll) emphakathini womthombo ovulekile we-TensorFlow wokusheshisa umsebenzi we-AI.
• I-Intel® Distribution yePython*: Thola ukusebenza kohlelo lokusebenza olusheshayo lwePython khona kanye ngaphandle kwebhokisi, ngoshintsho oluncane noma olungenalo kwikhodi yakho. Lokhu kusatshalaliswa kuhlanganiswe ne-Intel® Performance Libraries efana ne-Intel® oneAPI Math Kernel Library kanye ne-Intel®oneAPI Data Analytics Library.
• I-Intel® Distribution ye-Modin* (etholakala nge-Anaconda kuphela), ekwenza ukwazi ukukala kalula ukucutshungulwa kwangaphambili kumanodi amaningi usebenzisa lo mtapo wedatha ohlakaniphile, osabalalisiwe one-API efanayo kuma-panda. Lokhu kusatshalaliswa kutholakala kuphela Ngokufaka I-Intel® AI Analytics Toolkit nge-Conda* Umphathi Wephakheji.
• I-Intel® Neural Compressor : sebenzisa ngokushesha izixazululo ze-inference ezinemba kancane kuzinhlaka zokufunda ezijulile ezifana ne-TensorFlow*, PyTorch*, MXNet*, kanye ne-ONNX* (Open Neural Network Exchange) isikhathi sokusebenza.
• I-Intel® Extension ye-Scikit-learn*: Indlela engenamthungo yokusheshisa uhlelo lwakho lokusebenza lwe-Scikit-lear usebenzisa i-Intel® oneAPI Data Analytics Library (oneDAL).
  I-Patching scikit-learn iyenza ibe wuhlaka lokufunda lomshini olufaneleka kahle lokubhekana nezinkinga zempilo yangempela.
• I-XGBoost Optimized by Intel: Le phakheji eyaziwa kakhulu yomshini wokufunda ngezihlahla zesinqumo ezithuthukisiwe ihlanganisa ukusheshisa okungenamthungo, okokwehla kwezakhiwo ze-Intel® ukuze kusheshiswe kakhulu ukuqeqeshwa kwamamodeli futhi kuthuthukiswe ukunemba kwezibikezelo ezingcono.

Lungiselela Uhlelo Lwakho – Intel® AI Analytics Toolkit

Uma ungakayifaki I-AI Analytics Toolkit, bheka Ukufaka I-Intel® AI Analytics Toolkit. Ukuze ulungiselele isistimu yakho, setha okuguquguqukayo kwendawo ngaphambi kokuqhubeka.

 

Setha Okuguquguqukayo Kwemvelo Kokuthuthukiswa Kwe-CLI
Ukuze usebenze ku-Command Line Interface (CLI), amathuluzi kukhithi yamathuluzi ye-oneAPI alungiswa ngokusebenzisa
okuguquguqukayo kwemvelo. Ukusetha okuguquguqukayo kwemvelo ngokuthola iskripthi se-setvars:

Inketho 1: Umthombo we-setvars.sh kanye ngeseshini ngayinye
Umthombo we-setvars.sh njalo uma uvula iwindi elisha letheminali:

Ungathola umbhalo we-setvars.sh kufolda yempande yokufakwa kwakho kwe-oneAPI, okuvamise ukuthi /opt/intel/oneapi/ yokufakwa okubanzi kwesistimu kanye ne-~/intel/oneapi/ yokufakwa kwangasese.

Ukufakwa okubanzi kwesistimu (idinga amalungelo wezimpande noma we-sudo):

• . /opt/intel/oneapi/setvars.sh

Okokufakwa okuyimfihlo:

• . ~/intel/oneapi/setvars.sh

Inketho 2: Ukusethwa kwesikhathi esisodwa kwe-setvars.sh
Ukuze wenze imvelo isethelwe amaphrojekthi akho ngokuzenzakalelayo, faka umthombo womyalo
/setvars.sh kusikripthi sokuqalisa lapho sizocelwa khona ngokuzenzakalelayo (faka esikhundleni
ngendlela eya endaweni yakho yokufaka ye-oneAPI). Izindawo zokufaka ezizenzakalelayo yilezi /opt/
i-intel/oneapi/ yokufakwa okubanzi kwesistimu (idinga izimpande noma amalungelo e-sudo) kanye ne- ~/intel/oneapi/ ekufakweni okuyimfihlo.
Okwesiboneloample, ungakwazi ukwengeza umthombo /setvars.sh umyalo ku ~/.bashrc noma ~/.bashrc_pro yakhofile noma ~/.profile file. Ukuze wenze izilungiselelo zibe unomphela kuwo wonke ama-akhawunti ohlelweni lwakho, dala umbhalo we-.sh womugqa owodwa ohlelweni lwakho /etc/profile.d ethola i-setvars.sh (ukuthola imininingwane eyengeziwe, bheka imibhalo ye-Ubuntu kokuthi Okuguquguqukayo Kwendawo).

QAPHELA
Iskripthi se-setvars.sh singaphathwa kusetshenziswa ukumisa file, okuwusizo ikakhulukazi uma udinga ukuqalisa izinguqulo ezithile zamalabhulali noma umdidiyeli, kunokushintshela enguqulweni “yakamuva”. Ukuze uthole imininingwane eyengeziwe, bheka Ukusebenzisa Ukucushwa File ukuze Uphathe i-Setvars.sh.. Uma udinga ukumisa indawo kugobolondo elingelona ele-POSIX, bona Isethaphu Yokuthuthukiswa Kwendawo Ye-API ukuze uthole izinketho ezengeziwe zokucushwa.

Izinyathelo Ezilandelayo

• Uma ungasebenzisi i-Conda, noma uthuthukisela i-GPU, Yakha futhi Uqalise i-Sample Project.
• Kubasebenzisi be-Conda, qhubekela esigabeni esilandelayo.
• Ukuze uthuthuke ku-GPU, qhubekela kubasebenzisi be-GPU

I-Conda Environments kule Toolkit
Kunezindawo eziningi ze-conda ezifakwe ku-AI Kit. Indawo ngayinye ichazwe kulelithebula elingezansi. Uma usubeke okuguquguqukayo kwemvelo endaweni ye-CLI njengoba kuyalelwe ngaphambili, ungavula izindawo ezihlukene ze-conda njengoba kudingeka ngomyalo olandelayo:

• conda activate

Ukuze uthole ulwazi olwengeziwe, sicela uhlole indawo ngayinye ehlobene ne-Getting Started Sampixhunywe kulelithebula elingezansi.

Sebenzisa Umsebenzi We-Conda Clone ukuze Wengeze Amaphakheji Njengomsebenzisi Ongeyena Izimpande
Ikhithi yamathuluzi ye-Intel AI Analytics ifakwe kufolda ye-oneapi, edinga amalungelo ezimpande ukuze uyiphathe. Ungase ufise ukwengeza nokugcina amaphakheji amasha usebenzisa i-Conda*, kodwa awukwazi ukwenza kanjalo ngaphandle kokufinyelela kwezimpande. Noma, ungase ube nokufinyelela kwezimpande kodwa ungafuni ukufaka iphasiwedi yempande njalo uma wenza kusebenze i-Conda.

Ukuphatha indawo yakho ngaphandle kokusebenzisa ukufinyelela kwezimpande, sebenzisa ukusebenza kwe-Conda clone ukuze uhlanganise amaphakheji owadingayo kufolda engaphandle kwefolda /opt/intel/oneapi/:

1. Kusukela kuwindi letheminali efanayo lapho usebenzise khona i-setvars.sh, khomba izindawo ze-Conda kusistimu yakho:
   • uhlu lwe-conda env
     Uzobona imiphumela efana nalena:
2. Sebenzisa umsebenzi we-clone ukuze wenze imvelo ifane nefolda entsha. Ku-exampngaphansi, indawo entsha ibizwa ngokuthi usr_intelpython futhi indawo ehlanganiswayo ibizwa ngesisekelo (njengoba kukhonjisiwe esithombeni esingenhla).
   • conda dala -name usr_intelpython -clone base
     Imininingwane ye-clone izovela:

1. Yenza kusebenze indawo entsha ukuze unike amandla amandla okwengeza amaphakheji. conda sebenzisa usr_intelpython
2. Qinisekisa ukuthi indawo entsha iyasebenza. uhlu lwe-conda env
   Manje usungakwazi ukuthuthukisa usebenzisa imvelo ye-Conda ye-Intel Distribution ye-Python.
3. Ukwenza kusebenze imvelo ye-TensorFlow* noma ye-PyTorch*:

I-TensorFlow

• conda activate tensorflow

I-PyTorch

• conda activate pytorch

Izinyathelo Ezilandelayo

• Uma ungathuthukeli i-GPU, Yakha futhi Uqalise i-Sample Project.
• Ukuze uthuthuke ku-GPU, qhubekela kubasebenzisi be-GPU.

Abasebenzisi be-GPU
Kulabo abathuthuka ku-GPU, landela lezi zinyathelo:

Faka abashayeli be-GPU
Uma ulandele iziqondiso Kumhlahlandlela Wokufaka ukuze ufake Amashayeli e-GPU, ungeqa lesi sinyathelo. Uma ungakazifaki izishayeli, landela izikhombisi-ndlela ezikumhlahlandlela wokufaka.

Engeza Umsebenzisi Eqenjini Levidiyo
Ngomthwalo wekhompyutha wokubala we-GPU, abasebenzisi abangeyona impande (abavamile) ngokuvamile abanakho ukufinyelela kudivayisi ye-GPU. Qiniseka ukuthi ungeza abasebenzisi bakho abajwayelekile eqenjini levidiyo; uma kungenjalo, okombambambili okuhlanganiswe kwedivayisi ye-GPU kuzohluleka uma kwenziwa umsebenzisi ojwayelekile. Ukuze ulungise le nkinga, engeza umsebenzisi ongelona izimpande eqenjini levidiyo:

• sudo usermod -a -G ividiyo

Khubaza i-Hangcheck
Ngezinhlelo zokusebenza ezine-GPU esebenza isikhathi eside yokubala imisebenzi ezindaweni zomdabu, khubaza i-hangcheck. Lokhu akunconyelwe ekwenzeni izinto ezibonakalayo noma okunye ukusetshenziswa okujwayelekile kwe-GPU, okufana nokudlala amageyimu.

Umthwalo wokusebenza othatha ngaphezu kwamasekhondi amane ukuze izingxenyekazi zekhompuyutha ze-GPU zisebenze kuwumthwalo omude osebenzayo. Ngokuzenzakalelayo, izintambo ezingazodwana ezifaneleka njengemithwalo yemisebenzi ehlala isikhathi eside zibhekwa njengelengiswa futhi ziyanqanyulwa. Ngokukhubaza isikhathi sokuvala se-hangcheck, ungagwema le nkinga.

QAPHELA: Uma i-kernel ibuyekeziwe, i-hangcheck ivulwa ngokuzenzakalelayo. Qalisa inqubo engezansi ngemva kokubuyekezwa kwe-kernel ukuze uqinisekise ukuthi i-hangcheck ivaliwe.

1. Vula itheminali.
2. Vula i-grub file ku-/etc/default.
3. Esigangeni file, thola umugqa GRUB_CMDLINE_LINUX_DEFAULT=”” .
4. Faka lo mbhalo phakathi kwezingcaphuno (“”):
5. Qalisa lo myalo:
   sudo update-grub
6. Qalisa kabusha isistimu. I-Hangcheck isalokhu ivaliwe.

Isinyathelo Esilandelayo
Manje njengoba usuyimisile isistimu yakho, qhubekela kokuthi Yakha futhi Uqalise i-Sample Project.

Yakha futhi Ugijime i-Sampusebenzisa i-Command Line

Intel® AI Analytics Toolkit
Kulesi sigaba, uzosebenzisa iphrojekthi elula ethi "Sawubona Umhlaba" ukuze uzijwayeze nenqubo yokwakha amaphrojekthi, bese uzakhela eyakho iphrojekthi.

QAPHELA: Uma ungakayilungisi indawo yakho yokuthuthuka, iya kokuthi Lungiselela isistimu yakho bese ubuyela kuleli khasi. Uma usuziqedile izinyathelo zokumisa uhlelo lwakho, qhubeka nezinyathelo ezingezansi.

Ungasebenzisa iwindi letheminali noma Ikhodi ye-Visual Studio* lapho usebenza kulayini womyalo. Ukuze uthole imininingwane yokuthi ungasebenzisa kanjani Ikhodi ye-VS endaweni, bheka Ukusetshenziswa Okuyisisekelo Kwekhodi Yesitudiyo Esibonakalayo nge-OneAPI ku-Linux*. Ukuze usebenzise Ikhodi ye-VS ukude, bona Ukuthuthukiswa Kwekhodi Yesitudiyo Esikude nge-API eyodwa ku-Linux*.

Yakha futhi Ugijime i-Sample Project
I-sampOkuncane okungezansi kumele kwenziwe kusistimu yakho ngaphambi kokuthi wakhe i-sample phrojekthi:

Ukuze ubone uhlu lwezingxenye ezisekela i-CMake, bona Sebenzisa i-CMake kuya nge-OneAPI Applications.

Yakha Iphrojekthi Yakho
Azikho izinguquko ezikhethekile kumaphrojekthi akho e-Python adingekayo ukuze uqale ukuwasebenzisa ngale khithi yamathuluzi. Kumaphrojekthi amasha, inqubo ilandela eduze inqubo esetshenziselwa ukudala i-sampamaphrojekthi we-Hello World. Bheka ku-Hello World FUNDA files ukuze uthole imiyalelo.

Ukwandisa Ukusebenza
Ungathola imibhalo yokukusiza ukuthi ukhulise ukusebenza kwe-TensorFlow noma i-PyTorch.

Lungiselela Indawo Yakho

QAPHELA: Uma indawo yakho ebonakalayo ingatholakali, noma uma ufisa ukwengeza amaphakheji endaweni yakho ebonakalayo, qiniseka ukuthi uziqedile izinyathelo kokuthi Sebenzisa Umsebenzi We-Conda Clone Ukwengeza Amaphakheji Njengomsebenzisi Ongeyena Izimpande.

Uma uthuthuka ngaphandle kwesiqukathi, thola umbhalo olandelayo ukuze usebenzise i-Intel® Distribution yePython*:

• • /setvars.sh
• lapho yilapho ufake khona le khithi yamathuluzi. Ngokuzenzakalelayo inkomba yokufaka ithi:
• Ukufakwa kwezimpande noma kwe-sudo: /opt/intel/oneapi
• Ukufakwa komsebenzisi wendawo: ~/intel/oneapi

QAPHELA: Iskripthi se-setvars.sh singaphathwa kusetshenziswa ukumisa file, okuwusizo ikakhulukazi uma udinga ukuqalisa izinguqulo ezithile zamalabhulali noma umdidiyeli, kunokushintshela enguqulweni “yakamuva”. Ukuze uthole imininingwane eyengeziwe, bheka Ukusebenzisa Ukucushwa File ukuze Uphathe i-Setvars.sh. Uma udinga ukumisa imvelo kugobolondo elingelona ele-POSIX, bona i-API Yokusethwa Kwendawo Yokuthuthukisa ukuze uthole izinketho eziningi zokumisa.

Ukushintsha izindawo, kufanele uqale uvale indawo esebenzayo.
I-ex elandelayoample ikhombisa ukumisa imvelo, isebenze i-TensorFlow*, bese ibuyela ku-Intel Distribution yePython:

Landa Isitsha

Intel® AI Analytics Toolkit
Iziqukathi zikuvumela ukuthi usethe futhi ulungise izindawo zokwakha, ukusebenzisa kanye nokuphrofayilila izinhlelo zokusebenza ze-oneAPI futhi usabalalise usebenzisa izithombe:

• Ungafaka isithombe esiqukethe indawo elungiselelwe ngaphambilini nawo wonke amathuluzi owadingayo, bese uthuthuke kuleyo ndawo.
• Ungagcina indawo futhi usebenzise isithombe ukuhambisa leyo ndawo komunye umshini ngaphandle kokusetha okwengeziwe.
• Ungalungisa iziqukathi ezinamasethi ahlukene ezilimi nezikhathi zokusebenza, amathuluzi okuhlaziya, noma amanye amathuluzi, njengoba kudingeka.

Landa isithombe se-Docker*
Ungalanda isithombe se-Docker* ku-Containers Repository.

QAPHELA: Isithombe se-Docker singu-~5 GB futhi singathatha imizuzu engu-15 ukulanda. Kuzodinga u-25 GB wesikhala sediski.

1. Chaza isithombe:
   image=intel/oneapi-aikit docker donsa “$image”
2. Donsa isithombe.
   i-docker idonsa "isithombe se-$"

Uma isithombe sakho sesilandiwe, qhubekela ekusebenziseni Iziqukathi ezinomugqa womyalo.

Ukusebenzisa Iziqukathi ezinomugqa womyalo
I-Intel® AI Analytics Toolkit Landa iziqukathi ezakhelwe kusengaphambili ngokuqondile. Umyalo ongezansi we-CPU uzokushiya ngesikhathi somyalo, ngaphakathi kwesiqukathi, kumodi yokusebenzisana.

CPU
image=intel/oneapi-aikit docker run -it “$image”

Kusetshenziswa i-Intel® Advisor, Intel® Inspector noma i-VTune™ eneziqukathi
Lapho usebenzisa lawa mathuluzi, amakhono engeziwe kufanele anikezwe esitsheni: –cap-add=SYS_ADMIN -cap-add=SYS_PTRACE

• docker run -cap-add=SYS_ADMIN -cap-add=SYS_PTRACE \ -device=/dev/dri -it "$image"

Ukusebenzisa i-Cloud CI Systems

Amasistimu e-Cloud CI akuvumela ukuthi wakhe futhi uhlole isofthiwe yakho ngokuzenzakalelayo. Bona i-repo ku-github ngokwesiboneloampokuncane kokucushwa files abasebenzisa i-oneAPI yezinhlelo ze-CI zamafu.

Ukuxazulula izinkinga nge-Intel® AI Analytics Toolkit

Izaziso Nemishwana Yokuzikhipha Emthwalweni

Ubuchwepheshe be-Intel bungadinga ihadiwe enikwe amandla, isofthiwe noma isevisi isebenze. Awukho umkhiqizo noma ingxenye engavikeleka ngokuphelele.
Izindleko zakho nemiphumela ingahluka.

© Intel Corporation. I-Intel, ilogo ye-Intel, nezinye izimpawu ze-Intel yizimpawu zokuthengisa ze-Intel Corporation noma izinkampani ezingaphansi kwayo. Amanye amagama namabhrendi angafunwa njengempahla yabanye.

Ulwazi Lomkhiqizo Nokusebenza

Ukusebenza kuyehluka ngokusetshenziswa, ukumisa nezinye izici. Funda kabanzi ku www.Intel.com/PerformanceIndex.
Ukubuyekezwa kwesaziso #20201201

Ayikho ilayisensi (eshiwo noma eshiwo, nge-estoppel noma ngenye indlela) yanoma imaphi amalungelo empahla yengqondo enikezwa yile dokhumenti. Imikhiqizo echazwe ingase iqukathe ukonakala kwedizayini noma amaphutha aziwa ngokuthi i-errata angabangela umkhiqizo ukuthi uphambuke ekucacisweni okushicilelwe. Izinkinga zamanje ezinezimpawu ziyatholakala ngesicelo.

I-Intel iyazilahla zonke iziqinisekiso ezicacile nezicatshangelwayo, okuhlanganisa ngaphandle komkhawulo, iziqinisekiso ezishiwo zokuthengiselana, ukufaneleka kwenjongo ethile, nokungaphuli umthetho, kanye nanoma yisiphi isiqinisekiso esivela ngenxa yokusebenza, inkambo yokuthengisa, noma ukusetshenziswa kwezohwebo.

Amadokhumenti / Izinsiza

Intel AI Analytics Toolkit for Linux [pdf] Umhlahlandlela Womsebenzisi Ikhithi yamathuluzi e-AI Analytics ye-Linux, i-AI Analytics Toolkit, i-Analytic Toolkit ye-Linux, i-Analytic Toolkit, ikhithi yamathuluzi.

Izithenjwa

• Imaniwali yosebenzisayo