I have an outdated neural network training python2.7 script which utilizes keras 2.1 on top of tensorflow 1.4; and I want it to be trained on my nVidia GPU; and I have CUDA SDK 10.2 installed on Linux. I thought Docker Hub is exactly for publishing frozen software packages which just work, but it seems there is no way to find a container with specific software set.
I know docker >=19.3 has native gpu support, and that nvidia-docker utility has cuda-agnostic layer; but the problem is i cannot install both keras-gpu and tensorflow-gpu of required versions, cannot find wheels, and this legacy script does not work with other versions.
Asked
Active
Viewed 79 times
-3
1 Answers
-1
Where did you get the idea tha Docker Hub hosts images with all possible library combinations?
If you cannot find an image that suits you, simply build your own.
Konrad Botor
- 4,765
- 1
- 16
- 26
-
1) why would i need to containerize my software stack if i can build it? 2) as i already said, i cannot find tf-gpu1.4 wheel and cannot install it along keras-gpu2.1 with normal methods, like conda install 3) all these seems incompatible with cuda10.2 – ivan866 Aug 10 '20 at 12:28
-
11) You would containerize it to avoid installing the libs on the host, to avoid clashes with other projects and for reusability. 2) and 3) If it cannot be done without Docker, it cannot be done with Docker. – Konrad Botor Aug 10 '20 at 12:37
-
1) but in order to build my own docker image i HAVE to install libs on my host; 2) clashes are easily addressed with venv's; 3) if it WAS possible in 2014, it should be possible in 2020, given a cuda-agnostic gpu driver – ivan866 Aug 10 '20 at 12:47
-
21) No, you don't. You just add appropriate `RUN` statement to your Dockerfile and the packaes are isntalled inside the image only. 2) Yes, in Python's case there are mutliple ways to avoid clashes, Docker is just one of them. 3) Theoretically yes, but if the old versions of the libs are no logner available in the repos... – Konrad Botor Aug 10 '20 at 13:00