With our PyTorch image downloaded from NGC, we can now launch a container and investigate the contents. To view a full list of images installed, run docker images
.
On your workstation, launch the container while specifying that you want all available GPUs to be included. If you do not have an NVIDIA GPU and did not install the NVIDIA Container Toolkit, remove the --gpus all
flag from the docker run
command to launch without GPUs.
$ docker run --rm -it --gpus all nvcr.io/nvidia/pytorch:22.03-py3
=============
== PyTorch ==
=============
NVIDIA Release 22.03 (build 33569136)
PyTorch Version 1.12.0a0+2c916ef
Container image Copyright (c) 2022, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2014-2022 Facebook Inc.
Copyright (c) 2011-2014 Idiap Research Institute (Ronan Collobert)
Copyright (c) 2012-2014 Deepmind Technologies (Koray Kavukcuoglu)
Copyright (c) 2011-2012 NEC Laboratories America (Koray Kavukcuoglu)
Copyright (c) 2011-2013 NYU (Clement Farabet)
Copyright (c) 2006-2010 NEC Laboratories America (Ronan Collobert, Leon Bottou, Iain Melvin, Jason Weston)
Copyright (c) 2006 Idiap Research Institute (Samy Bengio)
Copyright (c) 2001-2004 Idiap Research Institute (Ronan Collobert, Samy Bengio, Johnny Mariethoz)
Copyright (c) 2015 Google Inc.
Copyright (c) 2015 Yangqing Jia
Copyright (c) 2013-2016 The Caffe contributors
All rights reserved.
Various files include modifications (c) NVIDIA CORPORATION & AFFILIATES. All rights reserved.
This container image and its contents are governed by the NVIDIA Deep Learning Container License.
By pulling and using the container, you accept the terms and conditions of this license:
https://developer.nvidia.com/ngc/nvidia-deep-learning-container-license
root@ca44795386ae:/workspace#
If you see a message similar to the above, you are now inside the container.
To familiarize yourself with the image, type ls at the prompt to list the current directories contents. The output should look something like the following:
$ ls
NVIDIA_Deep_Learning_Container_License.pdf README.md docker-examples examples tutorials
This will very likely look different from the working directory you were in prior to launching the container which indicates that we are inside the container and viewing contents specific to it as anticipated.
To further demonstrate the flexibility containers provide in terms of installing packages, let’s install a new package using Ubuntu’s package manager. In general, images available on NGC are based on the Ubuntu Operating System which might be different from the OS installed on your workstation.
To install a package, run the following inside the container (note that you are running as root by default and sudo
is not required. In fact, sudo
is not a valid command inside the container):
$ apt update
$ apt install -y htop
This installs the htop application - a commonly-used process viewer for Linux. Running htop
will verify that it was installed and is now usable inside the container. Press “q” on your keyboard to quit out of htop.
If htop was not installed on your workstation prior to running the container, it will remain uninstalled on the physical host (ie. outside of the running container). This is as intended as it allows us to configure a tailored environment for a specific application inside the container without introducing dependency conflicts on the workstation. This is useful when working on different machines as they will very likely have different versions or combinations of packages installed, causing potential conflicts. Installing specific packages inside the container ensures all systems will run the same setup regardless of what is installed on the physical host.
On the flipside, since we are running the base container from NGC, we will need to install applications manually every time we launch the container. To prevent this, we can create a custom image that includes all of our packages, code, settings, and other updates to match exactly what we need. This allows us to launch a container without making manual changes inside.
If desired, a container that has been modified at runtime can be saved as a new image by using the docker commit
command, though it is recommended to use Dockerfiles (more on this next) to have a documented approach to reproduce an image for others. If you would like to commit a container as a new image, first find the container name with docker ps
in a separate terminal.
$ docker ps
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
bfee26ccb86e nvcr.io/nvidia/pytorch:22.03-py3 "/opt/nvidia/nvidia_…" 4 seconds ago Up 3 seconds 6006/tcp, 8888/tcp elegant_goldstine
The container was randomly named “elegant_goldstine” by Docker. We will use this name in the next step, though yours will very likely be different.
To save the running container as an image, run the following while changing the names as needed:
$ docker commit elegant_goldstine custom_image:1.0.0
Once finished looking at the container, type exit
to get out of the container.