Enabling AI Safety with NeMo Guardrails and Generative AI Red-teaming & Assessment Kit (GARAK)

Duration: …

Introduction

Course Title: Enabling AI Safety with NeMo Guardrails and Generative AI Red-teaming & Assessment Kit (GARAK)

Description: This course provides a hands-on lab covering NVIDIA NeMo Guardrails and Garak (Generative AI Red-teaming & Assessment Kit) Evaluation Framework. This hands-on lab focuses on how to use NeMo Guardrails and Garak Evaluation Framework features within the Red Hat OpenShift AI (RHOAI) environment.

Objectives

On completing this course, you should be able to:

  • Deploy and test NeMo Guardrails by using only built-in detectors and no LLM calls.

  • Deploy NeMo Guardrails with an LLM for advanced guardrail capabilities including self-check rails.

  • Test model security against known attacks using tools like Nvidia Garak.

Prerequisites

This course assumes that you have the following prior experience:

  • A basic understanding of machine learning principles and workflows is required.

  • A basic understanding of Generative AI and Large Language Models (LLMs) is required.

  • Basic experience with Git is required.

  • Experience with containers and Red Hat OpenShift is recommended, or completion of the Red Hat OpenShift Developer II: Building and Deploying Cloud-native Applications (DO288) course.

Contributors

The PTL team acknowledges the valuable contributions of the following Red Hat associates:

  • Sarvesh Pandit (Principal Technical Training Developer)

  • Mac Misiura (Senior Machine Learning Engineer)

  • Cansu Kavili Oernek (Principal AI Platform Architect)

  • Anneli Sara Banderby (Senior AI Architect)

  • Noel O’Connor (Senior Principal Architect)

  • Trusty AI Engineering Team

  • Rutuja Deshmukh (Technical Training Developer)

  • Anna Swicegood (Learning Product Manager)