Step 1. Obtain the Google deep learning container URL.
Google provides pre-configured and optimized deep learning containers that can be used within LiCO. To view available container images, visit
a. Select a container for your DL framework, for example, GPU-enabled Tensorflow 13.
b. Find the image version you want to use (the one with the “latest” tag is recommended), click , and select the Show Pull Command option.
c. Copy the last part of the command shown on the displayed page, that is, platform-release/tf-gpu.1-13:latest in this example.

Step 2. Prepare a recipe file.
Create a new file named recipe.def, using the URL reference obtained in step 1:

Bootstrap: docker
chmod 755 /root

Step 3. Build a singularity image from the recipe file.
Build an image on the server with singularity by running the following command as a root user:

$ singularity build tf-gpu.1-13.image recipe.def

A container named tf-gpu-1.13.img is created.