Pull images from NVIDIA GPU CLOUD
To pull images from NVIDIA’s DockerHub repository, one needs to register and setup an API Key which will sit is a hidden file in one’s home directory. Every time we access NGC, this file will be queried by the image registry to verify identity.
Creating an account on NGC
Step 1
Go to https://www.nvidia.com/en-us/gpu-cloud/ and click on “Explore NGC Catalog”
Step 2
On the left top click on “Welcome Guest” and select “SignIn/SignOn”
This will take you to the following page where you select NVIDIA Account. If you have an existing NVIDIA developer account, use this otherwise create one:
Step 3
Once signed in, click the drop down menu on left top with your profile and select Setup. This should take you to the following page.
We have downloaded and installed the NGC CLI for you on Ibex as a module file. You can click on the Get API Key box and click “Generate API” on the subsequent page. Copy the API key you get
Please load the module by logging on to glogin node and invoke the following command:
module load ngc
ngc config setPaste the copied API key. For the org choices, copy the org name adjacent to your profile name on the NGC Setup webpage as in Fig 2 and paste in your terminal. For all other prompts, simple press enter and select the default values.
$ ngc config set
Enter API key [no-apikey]. Choices: [<VALID_APIKEY>, 'no-apikey']: *****
Enter CLI output format type [ascii]. Choices: [ascii, csv, json]:
Enter org [no-org]. Choices: ['******']:
Enter team [no-team]. Choices: ['no-team']:
Enter ace [no-ace]. Choices: ['no-ace']:
Successfully saved NGC configuration to /home/shaima0d/.ngc/config
Pulling an Image with Singularity
NGC container images are usually big in size. We prefer to pull these images either on WekaIO filesystem or home directory. First set the Singularity temporary directory to home
export SINGULARITY_TMPDIR=$HOME/tmpdirNow we pull the image we are interested in:
singularity pull docker://nvcr.io/nvidia/pytorch:22.08-py3This will take some time, and when completed, you will end up with and executable file pytorch_22.08-py3.sif
Happy hunting