TensorFlow 2.0 inosvika, rakavhurika sosi raibhurari yekudzidza kwemuchina

tf_logo

Mamwe mazuva apfuura vhezheni itsva yakapihwa yakakosha yemuchina wekudzidza chikuva TensorFlow 2.0, , que inopa kunze-kwe-kwe-bhokisi kuitiswa kweakasiyana akadzika emuchina ekudzidza algorithms, yakapusa programming interface yekuvaka mamodheru muPython uye yakaderera-chikamu chinongedzo cheC ++ icho chinokutendera iwe kudzora kuvaka uye kuitisa kwemakomputa grafiti.

Nzvimbo yacho yakatanga kugadzirwa neboka reGoogle Brain uye inoshandiswa neGoogle services yezwi kucherechedzwa, kuzivikanwa kwechiso mumifananidzo, sarudza kufanana kwemifananidzo, kusefa spam muGmail, sarudza nhau muGoogle News uye kuronga shanduro maererano nezvazvinoreva.

TensorFlow inopa raibhurari yemakomputa algorithms kunze-kwe-kwe-mubhokisi manhamba anoitwa kuburikidza nedatha kuyerera machati. Iwo maodhi mumagirafu akadaro anoita masvomhu mashandiro kana nzvimbo dzekupinda / dzekubuda, nepo micheto yegrafu inomiririra multidimensional data sets (tensors) inoyerera pakati penodhi.

Node dzinogona kupihwa kumacomputer zvigadzirwa uye kumhanya asynchronously, panguva imwe chete kugadzirisa ese akakodzera tensors panguva imwe chete, zvichikubvumidza iwe kuronga panguva imwe chete mashandiro enode mune neural network nekufananidza pamwe nehumwechete kupfura mauroni muuropi.

Dzakaparadzirwa michina yekufunda masisitimu inogona kuvakwa pane zvakajairika michina, nekuda kwerutsigiro rwakavakirwa muTensorFlow yekuwedzera komputa kune akawanda maCPU kana maGPU. TensorFlow inogona kumhanya pane akawanda maCUU uye maGPU (aine sarudzo yeCUDA yekuwedzeredza yechinangwa chekombuta pamakomputa ekugadzirisa zvikamu)

TensorFlow inowanikwa pane 64-bit Linux, macOS, uye mapuratifomu efoni anosanganisira Android uye iOS. Iyo kodhi yekodhi yakanyorwa muC ++ uye Python uye inogoverwa pasi peiyo Apache rezinesi.

Main nyowani maficha eTensorFlow 2.0

Nokuburitswa kweshanduro iyi nyowani yekutarisisa zvikweretesa kuve nyore uye nyore kushandisa, yakadaro ndiyo nyaya yekuvaka uye kudzidzisa mamodheru, imwe nyowani yepamusoro-soro Keras API yakafemerwa iyo inopa akati wandei sarudzo dzekupindirana kuvaka mamodheru (akateedzana, anoshanda, subclass) aine mukana wekuurayiwa kwavo nekukasira (pasina kwekutanga kuunganidzwa) uye neyakareruka nzira yekugadzirisa.

Yakawedzera tf.distribute.Streneo API yekuronga yakaparadzirwa modhi yekudzidzisas nekushandurwa kushoma kune iripo kodhi Pamusoro pekukwanisa kugovera kuverenga kune akawanda maGPU, pane rutsigiro rwekuyedza ruripo rwekuparadzanisa maitiro ekudzidza kune akawanda akazvimirira maprosesa uye kugona kushandisa gore TPU (Tensor Processing Unit).

Panzvimbo peyekuzivisa girafu modhi yekumisikidza nekuitwa kuburikidza tf.Session, zvinokwanisika kunyora zvakajairika Python mabasa ayo anogona kushandurwa kuita magirafu nekufonera tf. Kushanda uye wozoita kure kure, serialized kana kukwidziridzwa kusimudzira mashandiro. kuita.

Yakawedzerwa muturikiri weAutoGraph anoshandura Python kuraira kuyerera kuTensorFlow expression, iyo inokutendera iwe kushandisa Python kodhi mukati me tf.function, tf.data, tf.distribute, uye tf.keras mabasa.

SavedModel yakabatanidza iyo modhi yekuchinjana fomati uye yakawedzera rutsigiro rwekuchengetedza uye kudzoreredza mamiriro emhando Yakaunganidzwa mamodheru eTensorFlow ikozvino inogona kushandiswa muTensorFlow Lite (pane nhare mbozha), TensorFlow JS (mubrowser kana Node.js), TensorFlow Serving, uye TensorFlow Hub.

Iyo tf.train.Optimizers uye tf.keras.Optimizers APIs dzakabatana, Panzvimbo pekuti compute_gradients, nyowani GradientTape kirasi yakagadziridzwa kuenzanisa gradients.

Zvakare iko kuita mune iyi nyowani vhezheni kwave kwakanyanya kukwirira kana uchishandisa iyo GPU. Iko kumhanyisa kwekudzidzira modhi pane masisitimu ane NVIDIA Volta uye Turing GPUs yakawedzera kusvika katatu.

Yakawanda yekuchenesa APIs, mafoni mazhinji anotumidzwazve mazita kana kubviswa, tsigiro yemisiyano yepasi rose munzira dzekubatsira yaputswa. Panzvimbo ye tf.app, tf.flags, tf.logging, itsva absl-py API inokurudzirwa. Kuti uenderere mberi nekushandisa iyo yekare API, iyo compat.v1 module yakagadzirirwa.

Kana iwe uchida kuziva zvakawanda nezvazvo unogona kubvunza chinotevera chinongedzo.


Izvo zviri muchinyorwa zvinoomerera pamisimboti yedu ye tsika dzekunyora. Kuti utaure chikanganiso tinya pano.

Iva wekutanga kutaura

Siya yako yekutaura

Your kero e havazobvumirwi ichibudiswa.

*

*

  1. Inotarisira iyo data: Miguel Ángel Gatón
  2. Chinangwa cheiyo data: Kudzora SPAM, manejimendi manejimendi.
  3. Legitimation: Kubvuma kwako
  4. Kutaurirana kwedata
  5. Dhata yekuchengetedza: Dhatabhesi inobatwa neOccentus Networks (EU)
  6. Kodzero: Panguva ipi neipi iwe unogona kudzora, kupora uye kudzima ruzivo rwako

bool (chokwadi)