From the course: Microsoft Azure AI Essentials: Workloads and Machine Learning on Azure

Azure AI Foundry demo

From the course: Microsoft Azure AI Essentials: Workloads and Machine Learning on Azure

Azure AI Foundry demo

“

- [Instructor] Welcome to Azure AI Foundry. Let's get started by searching for Azure AI Foundry on Microsoft Bing. Once you've found the site, you should log in with your Azure credentials. We've already set up a project for this demo, but if you're starting fresh, you'll need to create a new one. In the model catalog, you can browse all available models. Let's pick the Phi-3 Medium model from Microsoft, which is a small language model. You can immediately try it out or deploy it based on your selection. After deployment, you'll find your deployed model s listed under the deployments tab. For example, I previously deployed a GPT-4 Omni model in an ADA embedding model. Now let's explore content filters. By default, filters are set to a medium level for categories like sexual, self-harm, harmful and violent content. I've configured stricter filters for my models in this example. I then applied these strict filters to my deployed models. With the models deployed and the filters in place, we now head to the project playground and try the chat feature. Here you can modify the system's message, which guides the model's behavior. Let's start by asking, what is the capital of Australia? And have the model create a two day itinerary for it. Then we adjust the system message to only allow answers about Japan. We then ask the same questions. Notice that the model now restricts its responses to Japan as instructed. You can also tweak other parameters like chat history and token limits, though that's beyond the demo scope. Now let's try using our own data. Imagine you have a travel goods company with product information in PDFs. Initially, the model won't know about the TrailMaster X4 Tent. So if you ask what is the price of the product, it might give incorrect or even fabricated answers. To address this, we upload your data via the add your data tab in the studio to build a RAG system, which you've covered previously. Azure AI search will automatically index your data making it searchable. For this demo, I'm using an index I created earlier. Once the data is connected, you can ask again about the price of the product, and this time the model will provide the correct answer with a citation. Let's also try a query about product warranty. Finally, deploying the setup is easy. Use the deploy button at the top to choose where you want to deploy it. A website, a Microsoft Teams app, or as a new Copilot in Copilot Studio.

Contents