Generative AI lab with Red Hat OpenShift AI

Introduction

Now that you’ve learned the basics of Red Hat OpenShift AI using a Predictive AI model, it’s time to explore Generative AI.
Generative AI is a subset of artificial intelligence that creates models that generate new data.
It’s used in a variety of applications, including image generation, text generation, and music generation. If you’ve used ChatGPT or StableDiffusion, you’ve used Generative AI.

In this lab, we will explore using a large language model (LLM) downloaded from Huggingface. We will be using a small LLM we can use with minimal resources. In fact, calling a "Small LLM" is a bit misleading, but "SLM" is not a thing, so…​🤷

In Generative AI, the early parts of the workflow are very different, compared to Predictive AI:

  • In predictive AI, you typically use historical data to train a new model from scratch.

  • In generative AI, the models have been trained for us already, and so we merely download them from a site like HuggingFace and then deploy them in our Model Serving UI.

So at a high level, we will:

  • Download a model from Huggingface

  • Deploy the model by using single-model serving with a serving runtime

  • Test the model API

About LLMs

Today, we will be using Google’s flan-t5-small because of compatibility, small size, and small requirements (no GPU). It is not a good model, but the process is the same for any of the compatible models.