ufoym / deepo Goto Github PK

Hi there, thanks for this great project.

I had an error running tensorflow with the GPU.
The error messages are like this:
E tensorflow/stream_executor/cuda/cuda_dnn.cc:396] Loaded runtime CuDNN library: 7102 (compatibility version 7100) but source was compiled with 7005 (compatibility version 7000). If using a binary install, upgrade your CuDNN library to match. If building from sources, make sure the library loaded at runtime matches a compatible version specified during compile configuration. F tensorflow/core/kernels/conv_ops.cc:712] Check failed: stream->parent()->GetConvolveAlgorithms( conv_parameters.ShouldIncludeWinogradNonfusedAlgo<T>(), &algorithms)
I can fix the issue by downloading an lower version of CuDnn and run the following command:
$ dpkg -i libcudnn7_7.0.5.15-1+cuda9.0_amd64.deb".

I just wonder can you fix this problem for the docker image?
Thanks.

I got unspecified error when trying to run opencv following this basic OpenCV getting started. The error is:

OpenCV(3.4.1) Error: Unspecified error (The function is not implemented. Rebuild the library with Windows, GTK+ 2.x or Carbon support. If you are on Ubuntu or Debian, install libgtk2.0-dev and pkg-config, then re-run cmake or configure script) in cvShowImage, file /root/opencv/modules/highgui/src/window.cpp, line 636
Traceback (most recent call last):
  File "image_get_started.py", line 8, in <module>
    cv2.imshow("image", img)
cv2.error: OpenCV(3.4.1) /root/opencv/modules/highgui/src/window.cpp:636: error: (-2) The function is not implemented. Rebuild the library with Windows, GTK+ 2.x or Carbon support. If you are on Ubuntu or Debian, install libgtk2.0-dev and pkg-config, then re-run cmake or configure script in function cvShowImage

I was using the keras-cpu-py36 while adding OpenCV in its Dockerfile. Looking back to OpenCV and the libgtk2.0-dev and pkg-config wasn't yet included.

Hello,

This docker image makes developing models so much easier! Is there any way to access the ML libraries installed with this docker image through an IDE such Spyder?

The jupyter notebook access is nice, but being able to run code in Spyder would make debugging a little bit easier.

Thanks in advance.

I keep having some random dependency issues when using Deepo. Is there any plans (or could there be) to move over to the Ananconda python distributions from Continuum analytics? I've never had dependency issues with their package management system.

Another one popular framework, specialized for ASR (audio speech recognition) is Kaldi

https://github.com/kaldi-asr/kaldi

Is there a reason why all the modules and other statements that are generated by the generator.py script are concatenated and listed as a single RUN statement in the final Dockerfile?

Especially for large images with long and error prone build processes this results in that caching can't be used at all. I already had many cases where I had to try different combinations of packages/libraries or implement new ones by myself and it always cost a hell of a time to have to rebuild the image all over again when a slight change around the end of Dockerfile had to be made.

What about changing the generator in such a way, that each module does not generate the final string to be added to the Dockerfile directly but just returns a list of strings, each being listed as a single RUN statement in the final Dockerfile?

This docker image is an excellent work. Thank you very much for the work.

Can I get older image with cuda 8.0 and cudnn v6.0 ?

Pytorch users need torchvision.

Thanks!

Hi, the issue I had is that the Jupyter Notebook is up and running, but I can't open any existing notebooks (blank) or create new ones (permission denied). The command I used was:

nvidia-docker run -d -p 6789:8888 \
-v /path/to/local/drive:/home/$USER \
--ipc=host \
--name Jupyter_$USER \
ufoym/deepo:all-jupyter-py36 jupyter notebook --no-browser --allow-root \
--ip=0.0.0.0 \
--NotebookApp.password='sha1:xxx'

Thanks in advance for your help!

Please add Deeplearning4J framework as well.

Thanks in advance! 😄

Please add python3-distutils:
+++ b/generator/modules/python.py @@ -34,6 +34,7 @@ class Python(Module): DEBIAN_FRONTEND=noninteractive $APT_INSTALL \ python3.6 \ python3.6-dev \ python3-distutils \ && \
Otherwise the container rocks.Good job and thanks for the effort.

Cheerz

Thank you for great Dockerfiles.

Small error found
Missing '' (backslash) in Dockerfile.all-jupyter-py27-cpu at line 307

sir:
I want to know the loction of the caffe and other sofewares.so that i can run the softwares withs samples.

Why is it Python 3.5 that is used and not 3.6?

When using the torch and loading images, it reports libjpeg not found.

Reproduce

$ th

require('image')
image.lena() or image.load('xxxx.jpg')

Output

/usr/local/share/lua/5.1/trepl/init.lua:389: module 'libjpeg' not found:No LuaRocks module found for libjpeg
	no field package.preload['libjpeg']
	no file '/root/.luarocks/share/lua/5.1/libjpeg.lua'
	no file '/root/.luarocks/share/lua/5.1/libjpeg/init.lua'
	no file '/usr/local/share/lua/5.1/libjpeg.lua'
	no file '/usr/local/share/lua/5.1/libjpeg/init.lua'
	no file './libjpeg.lua'
	no file '/usr/local/share/luajit-2.1.0-beta1/libjpeg.lua'
	no file '/.luarocks/share/lua/5.1/libjpeg.lua'
	no file '/.luarocks/share/lua/5.1/libjpeg/init.lua'
	no file '/root/.luarocks/lib/lua/5.1/libjpeg.so'
	no file '/usr/local/lib/lua/5.1/libjpeg.so'
	no file './libjpeg.so'
	no file '/usr/local/lib/lua/5.1/loadall.so'
	no file '/.luarocks/lib/lua/5.1/libjpeg.so'	
warning: <libjpeg> could not be loaded (is it installed?)	
/usr/local/share/lua/5.1/trepl/init.lua:389: module 'liblua_png' not found:No LuaRocks module found for liblua_png
	no field package.preload['liblua_png']
	no file '/root/.luarocks/share/lua/5.1/liblua_png.lua'
	no file '/root/.luarocks/share/lua/5.1/liblua_png/init.lua'
	no file '/usr/local/share/lua/5.1/liblua_png.lua'
	no file '/usr/local/share/lua/5.1/liblua_png/init.lua'
	no file './liblua_png.lua'
	no file '/usr/local/share/luajit-2.1.0-beta1/liblua_png.lua'
	no file '/.luarocks/share/lua/5.1/liblua_png.lua'
	no file '/.luarocks/share/lua/5.1/liblua_png/init.lua'
	no file '/root/.luarocks/lib/lua/5.1/liblua_png.so'
	no file '/usr/local/lib/lua/5.1/liblua_png.so'
	no file './liblua_png.so'
	no file '/usr/local/lib/lua/5.1/loadall.so'
	no file '/.luarocks/lib/lua/5.1/liblua_png.so'	
warning: <liblua_png> could not be loaded (is it installed?)	
/usr/local/share/lua/5.1/dok/inline.lua:738: <image.lena> no bindings available to load images (libjpeg AND libpng missing)
stack traceback:
	[C]: in function 'error'
	/usr/local/share/lua/5.1/dok/inline.lua:738: in function 'error'
	/usr/local/share/lua/5.1/image/init.lua:1690: in function 'lena'
	[string "_RESULT={image.lena()}"]:1: in main chunk
	[C]: in function 'xpcall'
	/usr/local/share/lua/5.1/trepl/init.lua:661: in function 'repl'
	/usr/local/lib/luarocks/rocks/trepl/scm-1/bin/th:204: in main chunk
	[C]: at 0x00405050	

Is it a docker image bug or a OS-related library missing? How do I fix it?

It looks like there is some package missing (python3-tk?), which prevents from normal usage of matplotlib.pyplot. Is there any workaround for this issue? Thanks!

Python 3.6.3 (default, Oct  6 2017, 08:44:35) 
[GCC 5.4.0 20160609] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import matplotlib.pyplot
Traceback (most recent call last):
  File "/usr/lib/python3.6/tkinter/__init__.py", line 37, in <module>
    import _tkinter
ModuleNotFoundError: No module named '_tkinter'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python3.6/dist-packages/matplotlib/pyplot.py", line 116, in <module>
    _backend_mod, new_figure_manager, draw_if_interactive, _show = pylab_setup()
  File "/usr/local/lib/python3.6/dist-packages/matplotlib/backends/__init__.py", line 60, in pylab_setup
    [backend_name], 0)
  File "/usr/local/lib/python3.6/dist-packages/matplotlib/backends/backend_tkagg.py", line 6, in <module>
    from six.moves import tkinter as Tk
  File "/usr/local/lib/python3.6/dist-packages/six.py", line 92, in __get__
    result = self._resolve()
  File "/usr/local/lib/python3.6/dist-packages/six.py", line 115, in _resolve
    return _import_module(self.mod)
  File "/usr/local/lib/python3.6/dist-packages/six.py", line 82, in _import_module
    __import__(name)
  File "/usr/lib/python3.6/tkinter/__init__.py", line 39, in <module>
    raise ImportError(str(msg) + ', please install the python3-tk package')
ImportError: No module named '_tkinter', please install the python3-tk package

:)

This is more of a general question, if you don't know off the top of your head please feel free to close. Is there any way to use deepo / nvidia-docker with a macOS machine running Docker client commands remotely connecting to a Linux docker-machine with nvidia-docker configured?

Thanks for your work! This looks great.

I was trying to convert the all-in-one to a Docker image, but I'm having issues with the cuda support. It converts OK (creates the image), but upon running it complains that it can't find cuda.

See below for the command to convert.
sudo singularity build py27-gpu.img docker://ufoym/deepo:all-py27

Any tips? It would be nice if this worked with Singularity as that is what is effectively used in HPC environments.

Hi,
Is it possible to integrate Jupyter notebook? Shall I work on this?

Hi, I am new to docker and there are still lots of things that I do not fully grasp. However, I managed to install the image with all the packets. However, I find that there are some python libraries could be helpful for me that are not installed. One way to overcome this is by manually installing those each time I run the docker image. However, I was wondering whether there is a better way to do it so that I do not have to constantly install the libraries.

Thanks in advance

Hi@ufoym Did the deepo:caffe2 include the Detectron module? I didn't find it after i pull this images.

When I use the command ' docker pull ufoym/deepo' and after the download for a few minute, it has the following tips: unauthorized: authentication required. Want to know how to fix it,thanks!!

I am trying to upgrade to tensorflow 1.4 but i have this error

ImportError: libcuda.so.1: cannot open shared object file: No such file or directory

Thanks for the great work! Love this repo. Would be even nicer if you could consider adding ONNX as well.

error pulling image configuration: Get https://dseasb33srnrn.cloudfront.net/re........................................: net/http: TLS handshake timeout

how to slove it? THX!

How to reproduce this issue?

git clone https://github.com/ufoym/deepo.git
cd deepo/docker
docker build -t ufoym/deepo:keras-py27 -f Dockerfile.keras-py27 .

Error log from docker build

• while $PIP_INSTALLing setuptools

Collecting setuptools
  Downloading setuptools-39.0.1-py2.py3-none-any.whl (569kB)
Collecting pip
  Downloading pip-10.0.0-py2.py3-none-any.whl (1.3MB)
Installing collected packages: setuptools, pip
  Found existing installation: pip 8.1.1
    Not uninstalling pip at /usr/lib/python2.7/dist-packages, outside environment /usr
Successfully installed pip-10.0.0 setuptools-39.0.1
Traceback (most recent call last):
  File "/usr/bin/pip", line 9, in <module>
    from pip import main
ImportError: cannot import name main

Thanks for the one-stop DL docker files. I'm using the all-jupyter-py36-cu90 and found the following errors when importing theano and lasagne.

import theano
/usr/local/lib/python3.6/dist-packages/theano/tests/main.py:6: DeprecationWarning: Importing from numpy.testing.nosetesteris deprecated, import from numpy.testing instead.
from numpy.testing.nosetester import NoseTester
Traceback (most recent call last):
File "", line 1, in
File "/usr/local/lib/python3.6/dist-packages/theano/init.py", line 156, in
import theano.gpuarray
File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/init.py", line 33, in
from . import fft, dnn, opt, extra_ops, multinomial, reduction, sort, rng_mrg, ctc
File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/fft.py", line 14, in
from .opt import register_opt, op_lifter, register_opt2
File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/opt.py", line 2801, in
from .dnn import (local_abstractconv_cudnn,
File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/dnn.py", line 339, in
handle_type = CUDNNDataType('cudnnHandle_t', 'cudnnDestroy')
File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/dnn.py", line 259, in CUDNNDataType
version=version(raises=False))
File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/dnn.py", line 319, in version
if not dnn_present():
File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/dnn.py", line 209, in dnn_present
dnn_present.avail, dnn_present.msg = _dnn_check_version()
File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/dnn.py", line 180, in _dnn_check_version
v = version()
File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/dnn.py", line 334, in version
"while the library is version %s." % v)
RuntimeError: Mixed dnn version. The header is version 7201 while the library is version 7102.

import lasagne
Traceback (most recent call last):
File "", line 1, in
File "/usr/local/lib/python3.6/dist-packages/lasagne/init.py", line 12, in
import theano
File "/usr/local/lib/python3.6/dist-packages/theano/init.py", line 144, in
if hasattr(theano.tests, "TheanoNoseTester"):
AttributeError: module 'theano' has no attribute 'tests'

Hi, I have a docker file which builds tensorflow from sources with latest cuda 9.2 and cudnn 7.1.4. Does this repo accept docker with source builds? I can send PR if you do! Thanks!

Needed for reinforcement learning research

Hi there,
I am a newbie who wants to tryout deepo image. Is this supposed to be run only on machine with Nvidia GPUs? Is there some way I can make it run on windows 10 or ubuntu-in-virtualbox on a machine with Intel HD graphics.

Why is the deepo-image tightly coupled to nvidia-docker (instead of just docker)?

Thanks,

• Rahul

Hey! I'm looking into what it would take to build deepo for cuda 10. I'm creating this issue as a somewhat of a placeholder, I'll put more details in the comments as I work through it. Is anyone already working on this?

Context: I'm one of the founders of vast.ai, a service that allows renting others' gpus, and we use docker; Deepo is one of our main suggested images. We have some folks with 2080 machines, but we're blocked on having image support for cuda 10 builds of the frameworks - eg, cuda 9 tf seems to break on cuda 10.

Thanks for the awesome images, btw :)

Wonderful deepo ! It made DL much more convenient. But I encountered some error while importing the framework MxNet.
I can import tensorflow, caffe, pytorch....But I can't import mxnet.
Python Interpreter told me : Illegal instruction(core dumped).
Have somebody else solved this problem before?

Your docker image is amazing and I have tested all deeplearning tools like caffe,tensorflow etc can work. But I want to find the location of the build folder of caffe.
Thanks.

Will ipython/jupyter be added in the future?

usage: ipykernel_launcher.py [-h] [--batch-size N] [--test-batch-size N]
[--epochs N] [--lr LR] [--momentum M] [--no-cuda]
[--seed S] [--log-interval N]
ipykernel_launcher.py: error: unrecognized arguments: -f /root/.local/share/jupyter/runtime/kernel-8d4d68c2-c1a5-4176-83d3-6d030e7d6e33.json

An exception has occurred, use %tb to see the full traceback.

SystemExit: 2

/usr/local/lib/python3.6/dist-packages/IPython/core/interactiveshell.py:2918: UserWarning: To exit: use 'exit', 'quit', or Ctrl-D.
warn("To exit: use 'exit', 'quit', or Ctrl-D.", stacklevel=1)

I want to run commond ifconfig and ip in the container, but 'apt install' doesn't work. Is there some wrong with package source?

I use image ufoym/deep:all-py27-jupyter

root@0767f14c50dc:/# apt install net-tools
Reading package lists... Done
Building dependency tree
Reading state information... Done
E: Unable to locate package net-tools

root@0767f14c50dc:/# apt install iputils-ping
Reading package lists... Done
Building dependency tree
Reading state information... Done
E: Unable to locate package iputils-ping

Thank you in advance ^_^

I'm running your all-py36 and all-py36-jupyter images on a remote, gpu-equipped machine. When I add -v ~/path/to/my/data:/data to the suggested nvidia-docker command for running the all-py36 image, I can access the data directory within the container. However, when I make the same addition to the suggested command for running the all-py36-jupyter image and then use the browser on my local machine to open the Jupyter Notebook at [SERVER ADDRESS]:[JUPYTER PORT], I don't see the data directory anywhere in the file menu. Am I doing something wrong?

It is my understanding that, in its current form, the container builds are not really reproducible as they use the most recent versions available. This is great in many cases, but undesirable in many others, e.g., if you want to build the exact same container that you would one year ago. Having explicit versions would mitigate this. Have you consider adding explicit version annotations in some form?

The deepo generator already supports plenty of common libraries and packages for deep learning and data science but in many cases one has a need for additional libraries that are most probably not supported. After implementing some common NLP libraries for the generator by myself (see #56, #57, #58), I had to realize that writing mostly redundant code for each pip or apt package doesn't really makes sense at all. Instead I think it would be better to support common package managers and enable to add any of such packages dynamically.

My idea is that with this approach packages that are not explicitly implemented could be still be added as follows:

python generator.py Dockerfile keras pip+pandas pip+nltk==3.3 apt+ping

Of course this wouldn't work out for ones that are difficult to install like OpenCV but at least it would give users an easy way to add any not explicitly supported packages.

I already started working on an implementation but currently have to rethink everything again as the current design of the generator and composer would probably not allow to support such a feature.

deepo seems very useful. I like the broad scope of supporting a diverse range of DL tools. But I believe it is missing one particular piece that I require. That piece is to build the C++ API, and is provided by https://github.com/FloopCZ/tensorflow_cc.

@FloopCZ

Hi, I had a question regarding the integration of a few features and plans for the future of this distribution. This Docker image has made my life soooo much easier, so thank you for that. I just had questions around if you ever plan to build in integration for Anaconda python or Jupyter lab? I was also wondering if you had a set release schedule for when you update the image with new Cuda distributions. Do you do that at certain intervals or just major releases, e.g. 8.0, 9.0 etc. Thanks again, Deepo has been a life saver.

Best,
Steve

Ubuntu 16.04 supplies Cmake 3.5.1-1ubuntu3, but Boost 1.65.1, installed in the Deepo containers, requires 3.9.3 or newer:

https://stackoverflow.com/questions/42123509/cmake-finds-boost-but-the-imported-targets-not-available-for-boost-version

CMake cannot detect the dependencies between the different Boost libraries. They have explicitly implemented in FindBoost.

For every Boost release this information is added by the CMake maintainers and it gets part of the next CMake release. So you have to make sure, that your CMake version was released after the Boost version you try to find.

Boost 1.63 requires CMake 3.7 or newer.
Boost 1.64 requires CMake 3.8 or newer.
Boost 1.65 and 1.65.1 require CMake 3.9.3 or newer.
Boost 1.66 requires CMake 3.11 or newer.
Boost 1.67 requires CMake 3.12 or newer.
Boost 1.68 and 1.69 require CMake 3.13 or newer.

Error occured when import theano in python

>>> import theano
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python3.6/dist-packages/theano/__init__.py", line 156, in <module>
    import theano.gpuarray
  File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/__init__.py", line 33, in <module>
    from . import fft, dnn, opt, extra_ops, multinomial, reduction, sort, rng_mrg, ctc
  File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/fft.py", line 14, in <module>
    from .opt import register_opt, op_lifter, register_opt2
  File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/opt.py", line 2801, in <module>
    from .dnn import (local_abstractconv_cudnn,
  File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/dnn.py", line 339, in <module>
    handle_type = CUDNNDataType('cudnnHandle_t', 'cudnnDestroy')
  File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/dnn.py", line 259, in CUDNNDataType
    version=version(raises=False))
  File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/dnn.py", line 319, in version
    if not dnn_present():
  File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/dnn.py", line 209, in dnn_present
    dnn_present.avail, dnn_present.msg = _dnn_check_version()
  File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/dnn.py", line 180, in _dnn_check_version
    v = version()
  File "/usr/local/lib/python3.6/dist-packages/theano/gpuarray/dnn.py", line 334, in version
    "while the library is version %s." % v)
RuntimeError: Mixed dnn version. The header is version 7104 while the library is version 7102.

Movidius is an neurochip device on an USB stick. Note: I am not an associate of Intel.
An example for gender detection (not my blog) https://jokersys.com/2017/08/05/first-look-movidius-neural-compute-stick/

when using python2.7 version deepo, I import caffe, and there is not Net and version module, is there any problem about this version?
Looking forward for your reply.
@ufoym