-
Notifications
You must be signed in to change notification settings - Fork 282
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
error run keras in docker #207
Comments
I can reproduce this as well with just this code: library(reticulate)
py_config() @eddelbuettel I saw there was a |
I discovered that the following additional apt-get dependencies are necessary to run tensorflow/keras in rocker: sudo apt-get install libpython2.7 python-pip python-virtualenv |
Hm, When you say Now we had good luck and adoption with the two containers for your two key products. Shall we look into a 'deep learning' rocker container with tf and keras? |
i use
would exist. i tried to build such an image but failed... |
@jjallaire
|
Yes I think a deep learning rocker container would be awesome! (esp. w/ the CUDA stuff pre-configured). The That said, calling Note that we still need |
I had to do |
Yes apt-get update && apt-get install ... or whichevre command you need. |
with
i get
with
i get the following error
so one step further but not done yet. any idea what i can do? |
Make sure cublas gets installed?
(That is on my 17.04 laptop) |
installed cublas with
and now get
hmm, i guess now i need to install libcusolver.so.8.0, right? thanks for your help! |
The people who maintain the rocker images aren't warranting that tensorflow will work out of the box. In this case I think you need to keep googling for the missing dependencies and then installing them manually (e.g. https://www.google.com/search?q=debian+libcusolver.so.8.0) |
And if you do have a list get in touch so that we may incorporate this. |
installed cublas with
get installed libcudart
get I and there i got stuck, could not find a way to install libcuda |
If you are attempting to install a GPU installation of TensorFlow on Ubuntu you should follow the documentation posted on the TensorFlow for R website to ensure you have the CUDA dependencies correctly installed an configured: https://tensorflow.rstudio.com/tools/installation_gpu.html#ubuntu |
i thought they should be all available since i use nvidia-docker, am i wrong? |
just to make sure i did run the nvidia-docker i rerun the installation
start bash shell in container install keras
install reticulate
same error message as last time, next step will be use the keras docker https://github.com/fchollet/keras/tree/master/docker and install RStudio there, i let you know how it went |
nvidia-docker as I understand simply makes the NVIDIA GPUs available the container. It doesn't do software installation or configuration as that is entirely application and library version specific. |
@jjallaire , i will try tomorrow to install the CUDA dependencies, my approach to run from the keras
then in shell
folllow RStudio instructions https://www.rstudio.com/products/rstudio/download-server/
and then i get an error i could not resolve:
that is as much as i can take for a weekend, i will continue tomorrow with starting from rocker/verse and adding CUDA dependencies correctly installed an configured: https://tensorflow.rstudio.com/tools/installation_gpu.html#ubuntu. |
ttried to install the tensorflow prerequisits in the container rocker/verse start bash shell in container install tensorflow prerequisits
i get
|
Try to install cuDNN
looks like the next dead end. |
at long last i got RStudio running with keras in GPU mode
that worked so, stop R and start RStudio
so it looks like it is working. |
@eddelbuettel Would you take a pull request for another rocker derived image as described here? |
I think we could do that, and as discussed, it work make a lot of sense. I am trying to remember if I have a box with NVidia somewhere to actually test this. This (from my standard desktop) should do, right? It is not a dedicated gpu card but driven by a multi-display nvidia. .....@.......:~$ lsmod | grep nvidia
nvidia_uvm 671744 0
nvidia_drm 45056 1
nvidia_modeset 843776 7 nvidia_drm
nvidia 13004800 626 nvidia_modeset,nvidia_uvm
drm_kms_helper 151552 1 nvidia_drm
drm 352256 4 nvidia_drm,drm_kms_helper
.....@.......:~$ |
issue closed, effort to create DL docker was started, see rocker-org/rocker#273 |
running keras in docker on Ubuntu 16.04 docker file https://hub.docker.com/r/rocker/verse/.
after installing keras as described on https://keras.rstudio.com when issuing the following command
> model <- keras_model_sequential()
i receive the error message
i would appreciate as to what i can do or if there is a dockerfile which avoids this problem.
thanks for your effort to make keras available for R, makes a big difference for me
The text was updated successfully, but these errors were encountered: