-1

I am new to docker. I am trying to run one github repository in docker container. I have few questions regarding the same.

  1. The docker image was created using CUDA version 9.0. I have Tesla T4 GPU and Driver version 470.129.06 in my amazon EC2 instance. I want to update CUDA version from 9 to 10.0. Is is possible to update cuda version inside the docker container? If yes, then How do we update it? Can you share some useful links?

  2. Whenever I run nvidia-docker run command it enters into fish interactive shell. I want it to start bash interactive shell. I don’t know how to enter in bash shell after running docker

The command I am running is:

nvidia-docker run -it --rm -v /Path_to_dataset:/root/data -v $Path_to_model:/root/model --ipc=host scrin/second-pytorch

I tried adding bash in command but does not work (I get error: open: No such file or directory).

nvidia-docker run -it --rm -v /Path_to_dataset:/root/data -v $Path_to_model:/root/model --ipc=host scrin/second-pytorch bash

The git repository I am using is: https://github.com/traveller59/second.pytorch

If anyone can help me. Thank you in advance.

Hetali
  • 43
  • 8
  • It sounds like you need to make some changes in your image's Dockerfile. Can you [edit] the question to include the source code necessary to reproduce your issue? – David Maze Jun 09 '22 at 10:25
  • I have added git repository link in the question. @DavidMaze – Hetali Jun 09 '22 at 11:40
  • 1
    You need to include a [mcve], including the required source code, in the question itself and not behind a link. (If you resolve the issue and update the GitHub repository, the link won't be especially meaningful.) [ask] has some more tips on writing a good question. – David Maze Jun 09 '22 at 11:48
  • 1
    Have you created the docker image yourself? Do you have full access to docker commands in your instance (not only running), you can copy/download bash binaries into your docker, can you manually update the CUDA version? – Sebastian Jun 09 '22 at 12:49
  • No, I haven't created image by myself. I am using the available image in hub.docker.com. The image I am using is "docker pull scrin/second-pytorch". I want to update CUDA manually. but not sure how to do it. @Sebastian – Hetali Jun 10 '22 at 06:30
  • Perhaps try to build the image yourself. On the linked Github there is a Dockerfile https://github.com/traveller59/second.pytorch/blob/master/Dockerfile which in turn refers to the base image nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04 That probably is the best start to change the CUDA version instead of updating afterwards (which you can also try, if changing the base image leads to nowhere). – Sebastian Jun 10 '22 at 08:22

1 Answers1

2

I found the solution.

1.

wget https://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/cuda_10.2.89_440.33.01_linux.run
sh cuda_10.2.89_440.33.01_linux.run
  1. If you enter in fish shell run command bash then it will enter in bash interactive shell.
Hetali
  • 43
  • 8