Docker 安装tensorflow

nvidia-dockerrun--rmnvidia/cuda:8.0-develnvidia-smi

3.InstallDockerandnvidia-docker

#Installdocker
curl-sSLhttps://get.docker.com/|sh

ThedockercontainerneedsaccesstotheGPUdevices.Forthispurposeuse`nvidia-docker`whichisawrapperaroundthestandard`docker`command.

#Installnvidia-dockerandnvidia-docker-plugin

wget-P/tmp'target='_blank'>https://github.com/NVIDIA/nvidia-docker/releases/download/v1.0.0-rc.3/nvidia-docker_1.0.0.rc.3-1_amd64.deb[/code]
sudodpkg-i/tmp/nvidia-docker*.deb&&rm/tmp/nvidia-docker*.deb

#Testnvidia-smi.
nvidia-dockerrun--rmnvidia/cudanvidia-smi

Youmightneedtouse`nvidia-docker`withsudo!

安装Tensorflow

http://blog.csdn.net/cq361106306/article/details/54094517
http://blog.csdn.net/freewebsys/article/details/70237003http://blog.csdn.net/dream_an/article/details/55520205
下载tensorflow-gpu版本的源
dockerpulldaocloud.io/daocloud/tensorflow:0.11.0-gpu

先查下你有几个GPU设备

[root@XXX~]#ls-la/dev|grepnvidia
crw-rw-rw-.1rootroot195,0Sep1613:49nvidia0
crw-rw-rw-.1rootroot195,255Sep1613:49nvidiactl
crw-rw-rw-.1rootroot247,0Sep1613:54nvidia-uvm

然后再查你的docker镜像

y@y:~$sudodockerimages
[sudo]passwordfory:
REPOSITORYTAGIMAGEIDCREATEDSIZE
daocloud.io/daocloud/tensorflow0.11.0-gpudd645f420f1d8weeksago2.713GB
daocloud.io/daocloud/tensorflow0.10.0-devel-gpufa886c09638d3monthsago5.014GB
hello-world

然后就可以启动咯

sudodockerrun-ti-v/home/:/mnt/home--privileged=true--device/dev/nvidia0:/dev/nvidia0--device/dev/nvidiactl:/dev/nvidiactl--device/dev/nvidia-uvm:/dev/nvidia-uvmdaocloud.io/daocloud/tensorflow:0.11.0-gpu/bin/bash



上面这句有点长把它写到docker.sh文件，然后

shdocker.sh

完成。上面的意思是把本地的/home映射到docker的/mnt目录
以及各种显卡设备也映射进去


主机保存镜像为新版本
sudodockerps-l
y@y:~$sudodockerps-l
CONTAINERIDIMAGECOMMANDCREATEDSTATUSPORTSNAMES
a1f2ac36a2c9daocloud.io/daocloud/tensorflow:0.11.0-gpu"/bin/bash"10minutesagoUp10minutes6006/tcp,8888/tcp

把a1f2ac36a2c9这个名字记住
然后

dockercommita1f2ac36a2c9新名字

OK了

1.https://medium.com/@gooshan/for-those-who-had-trouble-in-past-months-of-getting-google-s-tensorflow-to-work-inside-a-docker-9ec7a4df945b
4.RunaTensorflowGPU-enableDockercontainer
Thecontaineritselfisstartedaspointedoutintheofficialdocumentationasfollows:

#Runcontainer
nvidia-dockerrun-d--name<somename>-p8888:8888-p6006:6006gcr.io/tensorflow/tensorflow:latest-gpu

#Login
nvidia-dockerexec-it<somename>bash

e.g.:

nvidia-dockerrun-d--nametf1-p8888:8888-p6006:6006gcr.io/tensorflow/tensorflow:latest-gpu

nvidia-dockerexec-ittf1bash

Note:Port8888isforipythonnotebooksandport6006isforTensorBoard.

YoucantestifeverythingisalrightbyrunningthisPythonscript.

'target='_blank'>https://www.tensorflow.org/install/install_linux#InstallingDocker
GPUsupport

PriortoinstallingTensorFlowwithGPUsupport,ensurethatyoursystemmeetsallNVIDIAsoftwarerequirements.TolaunchaDockercontainerwithNVidiaGPUsupport,enteracommandofthefollowingformat:

$nvidia-dockerrun-it-phostPort:containerPortTensorFlowGPUImage

where:

-phostPort:containerPortisoptional.IfyouplantorunTensorFlowprogramsfromtheshell,omitthisoption.IfyouplantorunTensorFlowprogramsasJupyternotebooks,setbothhostPortand
containerPort
to
8888
.

TensorFlowGPUImagespecifiestheDockercontainer.Youmustspecifyoneofthefollowingvalues:

gcr.io/tensorflow/tensorflow:latest-gpu,whichisthelatestTensorFlowGPUbinaryimage.

gcr.io/tensorflow/tensorflow:latest-devel-gpu,whichisthelatestTensorFlowGPUBinaryimageplussourcecode.

gcr.io/tensorflow/tensorflow:version-gpu,whichisthespecifiedversion(forexample,0.12.1)oftheTensorFlowGPUbinaryimage.

gcr.io/tensorflow/tensorflow:version-devel-gpu,whichisthespecifiedversion(forexample,0.12.1)oftheTensorFlowGPUbinaryimageplussourcecode.

Werecommendinstallingoneofthe
latest
versions.Forexample,thefollowingcommandlaunchesthelatestTensorFlowGPUbinaryimageinaDockercontainerfromwhichyoucanrunTensorFlowprogramsinashell:

$nvidia-dockerrun-itgcr.io/tensorflow/tensorflow:latest-gpubash

ThefollowingcommandalsolaunchesthelatestTensorFlowGPUbinaryimageinaDockercontainer.InthisDockercontainer,youcanrunTensorFlowprogramsinaJupyternotebook:

$nvidia-dockerrun-it-p8888:8888gcr.io/tensorflow/tensorflow:latest-gpu

ThefollowingcommandinstallsanolderTensorFlowversion(0.12.1):

$nvidia-dockerrun-it-p8888:8888gcr.io/tensorflow/tensorflow:0.12.1-gpu

DockerwilldownloadtheTensorFlowbinaryimagethefirsttimeyoulaunchit.FormoredetailsseetheTensorFlowdockerreadme.

NextSteps
Youshouldnowvalidateyourinstallation.

使用：Docker
http://songhuiming.github.io/pages/2017/02/25/an-zhuang-dockerhe-tensorflow/
4.镜像管理
4.0.下载镜像

dockerpullimage-name


4.1.查看本地镜像

dockerimages


4.2.查看运行的容器

查看active镜像：dockerps
查看所有镜像：dockerps-a
查看最近的镜像：dockerps-l


4.3.删除镜像删除容器

dockerrmiimage-name

dockerrm<containerid>


4.4.在repo里搜索镜像

dockersearch[image-name]


4.5.停止镜像

dockerstopcontainer-id


4.6.停止并删除所有容器

dockerstop$(dockerps-a-q)
dockerrm$(dockerps-a-q)

5.4.退出

exit


Torestarttheexitedcontainer:

dockerstart-a-i`dockerps-q-l`


dockerstartstartacontainer(requiresnameorID)
-aattachtocontainer
-iinteractivemode
dockerpsListcontainers
-qlistonlycontainerIDs
-llistonlylastcreatedcontainer


5.5.HowdoIinstallnewlibrariesinDocker?

Therrearetwowaystodothis:

Firstmethodd:Modifythe
Dockerfile
directlytoinstallneworupdateyourexistinglibraries.Youwillneedtodoa
dockerbuild
afteryoudothis.IfyoujustwanttoupdatetoanewerversionoftheDLframework(s),youcanpassthemas
CLIparameter
usingthe
--build-argtag
(seefordetails).Theframeworkversionsaredefinedatthetopofthe
Dockerfile
.Forexample,
dockerbuild-tshmhub/dl-docker:cpu-fDockerfile.cpu--build-argTENSORFLOW_VERSION=1.2.0.


Secondmthodd:youcaninstallorupgradeinthecontainer.Afteritisdone,exitthecocntaineranddoa
commit
asintroductebelow.

5.5.向dockerimage提交containerchange

dockercommit-m"installvimwgetonubuntu"-a"author:shm"7de2c97f7a85shm/ubuntu_custom


这里7de2c97f7a85是imageid.也就是在bash下面看到的root@后面的id:
root@7de2c97f7a85
.这时候比较前后的镜像，就会发现commit以后多了一个镜像

shm@ubuntu:~$dockerimages
REPOSITORYTAGIMAGEIDCREATEDSIZE
tensorflow/tensorflowlatestea40dcc457242weeksago1.03GB
ubuntulatestf49eec89601e5weeksago129MB

shm@ubuntu:~$dockercommit-m"installvimwgetonubuntu"-a"author:shm"7de2c97f7a85shm/ubuntu_custom
sha256:5ed742f690e11c65db83936847c7c5659c5834f6b2c93b52d110455936e6a224

shm@ubuntu:~$dockerimages
REPOSITORYTAGIMAGEIDCREATEDSIZE
shm/ubuntu_customlatest5ed742f690e112secondsago647MB
tensorflow/tensorflowlatestea40dcc457242weeksago1.03GB
ubuntulatestf49eec89601e5weeksago129MB


5.6.向仓库提交镜像

首先登陆

dockerlogin-udocker-username


然后push

5.6.1.listtheimageandgetthetagid

shm@shm-xps9550:~/projects/dl_lessons/courses-master/deeplearning1/nbs$`dockerimages`
REPOSITORYTAGIMAGEIDCREATEDSIZE
shmhub/dl-dockercpu0f1e40d1bed812daysago9.13GB
ubuntu16.046a2f32de169d13daysago117MB


5.6.2.tagthegiagewiththe
registoryhost


dockertag0f1e40d1bed8pinseng/dl-docker


5.6.3.pushtheimagetotherepo

dockerpushdocker-username/docker-image-name
dockerpushpinseng/dl-docker


四、使用dockerexec进入Docker容器
除了上面几种做法之外，docker在1.3.X版本之后还提供了一个新的命令exec用于进入容器，这种方式相对更简单一些，下面我们来看一下该命令的使用：

[plain]viewplaincopy

$sudodockerexec--help



接下来我们使用该命令进入一个已经在运行的容器

[plain]viewplaincopy

$sudodockerps

$sudodockerexec-it775c7c9ee1e1/bin/bash



如何获取localhost的地址？
打开一个新的terminal，查看container的地址：

sudodockerinspectclever_bohr|grepIPAddress


1

这里的clever_bohr为该正在运行的container的名字，例子如下



在浏览器中输入：172.17.0.6:8888
http://blog.csdn.net/tina_ttl/article/details/51417358http://[allipaddressesonyoursystem]:8888/==========================>http://172.17.0.2:8888/



=========================================================================================================================================================================================================2017/11/8更=====

dockerpulltensorflow/tensorflow:nightly-gpu-py3



======================================dockerrunbash版本========================================



===========================================================本地ipython3notebook=================================================



=======================================================makepassword==================================================



========================================================usedocker-tfjupyter.shcode========================================================