Maalmo ka hor nooc cusub ayaa la soo bandhigay muhiim u ah mashiinka barashada mashiinka - TensorFlow 2.0, in waxay bixisaa hirgelinta sanduuqa-sanduuqa qaabab kala duwan oo ah algorithms-ka barashada mashiinka qoto dheer, interface interface fudud oo loogu talagalay qaababka dhismaha ee Python iyo dhexdhexaad heer hoose ah oo loo yaqaan 'C ++' kaas oo kuu oggolaanaya inaad xakameyso dhismaha iyo fulinta jaantusyada xisaabinta.
Goobta markii hore waxaa soo saaray kooxda Google Maskaxda waxaana adeegsada adeegyada Google aqoonsiga codka, aqoonsiga wajiga ee sawirada, go'aamiya isku ekaanshaha sawirada, shaandhada spam ee Gmail, xulo wararka kujira Google News iyo abaabul tarjumaada sida macnaha ah.
TensorFlow waxay bixisaa maktabad algorithms kombiyuutar lambar-ka-sanduuqa ka baxsan oo lagu hirgeliyey jaantusyada socodka xogta. Noodhadhka ku jira garaafyada noocan oo kale ah waxay hirgeliyaan hawlgallo xisaabeed ama dhibco gelitaan / bixid, halka cidhifyada garaafku ay metelaan xog uruurinno kala duwan (tensors) oo dhexmara qanjidhada.
Noodhadhka waxaa loo xilsaari karaa aaladaha kumbuyuutareedka isla markaana si isku mid ah u shaqeynaya, isla mar ahaantaana ka wada shaqeynaya dhammaan dareemayaasha ku habboon isla mar ahaantaana, taas oo kuu oggolaaneysa inaad abaabusho hawlgal isku mid ah oo qanjidhada ah oo ku jira shabakad neerfaha ah adigoo isbarbardhigaya isla mar ahaantaana tooganaya neerfaha maskaxda ku jira.
Nidaamyada barashada mashiinka loo qaybiyey waxaa lagu dhisi karaa qalabka caadiga ah, Thanks to dhisay-in support in TensorFlow for computing compute to multiple CPUs or GPUs. TensorFlow wuxuu ku shaqeyn karaa dhowr CPU iyo GPUs (oo leh ikhtiyaar kordhin CUDA ah oo loogu talagalay ujeedo guud oo xisaabeed oo ku saabsan cutubyada farsamada)
TensorFlow waxaa laga heli karaa 64-bit Linux, macOS, iyo barnaamijyada moobiilka oo ay ku jiraan Android iyo iOS. Nidaamka nambarka waxaa lagu qoray C ++ iyo Python waxaana lagu qaybiyaa liisanka Apache.
Astaamaha ugu muhiimsan ee cusub ee TensorFlow 2.0
Iyadoo la sii daayo nuucaan cusub dareenka ugu weyn amaahday si loo fududeeyo loona fududeeyo isticmaalka, sidan oo kale waa in la dhiso oo la tababaro moodooyinka, cusub oo heer sare ah Keras API ayaa la soo jeediyay kaas oo bixiya dhowr ikhtiyaar oo loogu talagalay isdhexgalka si loo dhiso moodello (isdaba-joog ah, shaqeynaya, subclass) iyada oo suurtagal ah in si dhakhso leh loo fuliyo (iyada oo aan la soo uruurin horudhac ah) iyo qaab fudud oo lagu jajabiyo.
Lagu daray tf.distribute.Strategy API si loo abaabulo tababarka qaabka loo qaybiyeys leh wax ka badal ugu yar ee koodhka jira. Marka lagu daro awoodda loo qaybiyo xisaabinta GPU-yo badan, waxaa jira taageero tijaabo ah oo loo heli karo in loo kala qaybiyo hannaanka wax-barasho ee processor-yo badan oo madax-bannaan iyo awoodda isticmaalka daruur TPU (Tensor Processing Unit).
Halkii laga qaadan lahaa qaab dhismeedka garaafka ee muujinta leh fulinta iyadoo loo marayo tf.Session, waxaa suurtagal ah in la qoro shaqooyinka caadiga ah ee loo yaqaan 'Python function' oo loo rogi karo garaafyada adoo wacaya tf.function ka dibna meel fog laga fulinayo, si taxaddar leh ama loo hagaajinayo waxqabadka. waxqabadka.
Ayaa lagu daray tarjumaan AutoGraph ah oo u rogaya qulqulka amarka Python ereyada TensorFlow, taas oo kuu oggolaaneysa inaad ku dhex isticmaasho koodhka Python-ka tf.function, tf.data, tf.distribute, iyo tf.keras.
SavedModel ayaa mideeyay qaabka isdhaafsiga moodeelka waxayna ku dartay taageerada keydinta iyo soo celinta gobolka moodooyinka. Moodooyinka la soo uruuriyay ee TensorFlow ayaa hadda loo isticmaali karaa TensorFlow Lite (aaladaha mobilada), TensorFlow JS (biraawsarka ama Node.js), TensorFlow Serving, iyo TensorFlow Hub.
Tf.train.Optimizers iyo tf.keras.Optimizers APIs ayaa la mideeyay, Halkii laga sameyn lahaa xisaabinta, fasal cusub oo loo yaqaan 'GradientTape' ayaa loo soo jeediyay in lagu xisaabiyo gradients.
Sidoo kale waxqabadka qaabkan cusub ayaa si aad ah uga sarreeyay markii la isticmaalayay GPU. Xawaaraha tababarka qaabka ee nidaamyada leh NVIDIA Volta iyo Turing GPUs ayaa kordhay illaa saddex jeer.
Qaar badan oo ka mid ah APIs nadiifinta, wicitaano badan ayaa dib loo beddelay ama laga saaray, taageerada isbeddelada caalamiga ah ee hababka caawinta ayaa jaban. Halkii laga heli lahaa tf.app, tf.flags, tf.logging, API cusub oo loo yaqaan 'absl-py API' ayaa la soo jeediyay. Si aad u sii wadato adeegsiga API-da hore, moduleka compres.v1 ayaa la diyaariyay.
Haddii aad rabto inaad wax badan ka ogaato waad la tashan kartaa xiriirka soo socda.