PyTorch for AI Workbench

PyTorch - NVIDIA AI Workbench Default Container (Beta)

NVIDIA AI Workbench provides a set of default base images, referred to as Base Environments in the application, that are used as the starting point when creating the container for a new project.

This is a PyTorch image with PyTorch 2.1.0 and CUDA 12.2 and with JupyterLab and Tensorboard applications installed that is configured for use with AI Workbench. It is based off of https://catalog.ngc.nvidia.com/orgs/nvidia/containers/pytorch.

AI Workbench is currently in Early Access. To access the software and start using this conatiner, first sign up here: https://developer.nvidia.com/ai-workbench-early-access

Use of this container falls under the AI Workbench EULA, which is available in DevZone after acceptance into the Early Access program.

CHANGELOG

All notable changes to this image are documented here

[1.0.1] - 2023-10-18

• First draft of image, based on nvidia/pytorch:23.09-py3
• Installed git-lfs
• Upgraded Jupyterlab to 4.0.7

[1.0.2] - 2023-11-13

• Added sources for apt packages

Shared Responsibility Model

NVIDIA AI Workbench employs a shared responsibility model. The subset of responsibilities related to base images are:

NVIDIA Responsibilities

1. Updating base environment images with a newer tag in case a new critical or high CVE is discovered.
2. Providing prompt notifications in the case of updates via
   • Notifications posted on the program DevZone site
   • Email to the accounts registered in the Early Access program on DevZone

End User and Organization Responsibilities

1. Following notifications from NVIDIA, ensuring the Default Base Images are updated to the latest versions to mitigate risks associated with vulnerabilities that might impact older images.