Exercise - Install the .NET SDK for Azure OpenAI and create your first app

  • 10 minutes

Now that we have an understanding of text and chat completions, let's create a basic application that uses them.

You work for an organization that helps hikers explore the outdoors by recommending trails. You're considering adding a conversational AI to the recommendation functionality and would like to create a prototype.

You decide you need to use the Chat Completion API provided by the GPT-35-Turbo model.

Create the Azure OpenAI resource

The first step is to create the Azure OpenAI resource and deploy the model. Let's use the GPT-3 Turbo model in this exercise. As the documentation notes, GPT-3.5 Turbo uses the Chat Completions API and is great for our use case.

Prerequisites

  • An Azure subscription

  • Access granted to Azure OpenAI in the desired subscription

    Currently, access to this service is granted only by application. You can apply for access to Azure OpenAI by completing the form at https://aka.ms/oai/access.

  • Azure CLI installed

Create a resource and deploying a model

Creating the resource and deploying the model is a multi-step process. Use the Azure CLI as it can be quicker than using the Azure portal. But note you can use the Azure portal if you wish.

  1. Run the az login command to sign-in if you haven't already.

  2. When you create a new Azure resource, you can create a new resource group or use an existing one. This command shows you how to create a new resource group. Use the name HikingConversations-RG but you can substitute the name of your choice, or use an existing group's name.

    az group create \
--name HikingConversations-RG \
--location eastus

    az group create `
--name HikingConversations-RG `
--location eastus

  3. Run the following command to create an OpenAI resource in the HikingConversations-RG resource group. Name the OpenAI resource HikingConversationsAI.

    az cognitiveservices account create \
-n HikingConversationsAI \
-g HikingConversations-RG \
-l eastus \
--kind OpenAI \
--sku s0

    az cognitiveservices account create `
-n HikingConversationsAI `
-g HikingConversations-RG `
-l eastus `
--kind OpenAI `
--sku s0

  4. Next we want to deploy the GPT-35-Turbo model to the OpenAI resource we created. Call the model deployment HikingRecommendationTurbo. Note we're using HikingConversations-RG as the resource group name, HikingConversationsAI as the OpenAI resource name, if you used different values make sure you substitute those values.

    az cognitiveservices account deployment create \
-g HikingConversations-RG \
-n HikingConversationsAI \
--deployment-name HikingRecommendationTurbo \
--model-name gpt-35-turbo \
--model-version "0301" \
--model-format OpenAI \
--scale-settings-scale-type "Standard"

    az cognitiveservices account deployment create `
-g HikingConversations-RG `
-n HikingConversationsAI `
--deployment-name HikingRecommendationTurbo `
--model-name gpt-35-turbo `
--model-version "0301" `
--model-format OpenAI `
--scale-settings-scale-type "Standard"

  5. Once the resource and model have been created, we need to get the base URL and access keys so the .NET SDK can access the Azure OpenAI resource. Use these commands to get the endpoint and primary API keys and make note of them for later use:

    The endpoint

    az cognitiveservices account show \
-g HikingConversations-RG \
-n HikingConversationsAI \
--query "properties.endpoint"

    az cognitiveservices account show `
-g HikingConversations-RG `
-n HikingConversationsAI `
--query "properties.endpoint"

    The primary API key

    az cognitiveservices account keys list \
-g HikingConversations-RG \
-n HikingConversationsAI \
--query "key1"

    az cognitiveservices account keys list `
-g HikingConversations-RG `
-n HikingConversationsAI `
--query "key1"

Create the console application and add the OpenAI SDK

Next up we want to create a bare bones .NET Console application and add the Azure OpenAI SDK.

  1. Run the following to create a new .NET application named HikingConversationsAI.

    dotnet new console -n HikingConversationsAI

  2. Switch to the newly created HikingConversationsAI directory.

    cd HikingConversationsAI

  3. Then add the Azure Open AI SDK.

    dotnet add package Azure.AI.OpenAI --prerelease

  4. Open the project in VS Code or Visual Studio.

  5. In the Program.cs file, delete all the existing code.

  6. Add using Azure.AI.OpenAI; to the top of Program.cs.

  7. Add using Azure; to the top of Program.cs.

  8. Add in three class-level variables that hold references to the Azure OpenAI resource's endpoint, the primary API key, and the name of the model you deployed.

    string openAIEndpoint = "<YOUR ENDPOINT URL VALUE>";
string openAIAPIKey = "<YOUR PRIMARY API KEY VALUE>";
string openAIDeploymentName = "HikingRecommendationTurbo";

    In the steps above, we named the deployment HikingRecommendationTurbo, if you used a different value make sure you use that instead.

  9. Finally, instantiate the class needed to communicate with the Azure OpenAI resource.

    var endpoint = new Uri(openAIEndpoint);
var credentials = new AzureKeyCredential(openAIAPIKey);
var openAIClient = new OpenAIClient(endpoint, credentials);

Create the system prompt

Let's create the initial system role prompt that will provide the initial instructions to the model.

  1. First create a ChatCompletionsOptions object which will define the parameters of how we want the model to react during the conversation and hold all of the messages in the conversation.

    var completionOptions = new ChatCompletionsOptions
{
    MaxTokens=400,
    Temperature=1f,
    FrequencyPenalty=0.0f,
    PresencePenalty=0.0f,
    NucleusSamplingFactor = 0.95f, // Top P
    DeploymentName = openAIDeploymentName
};

  2. Create the prompt that instructs the model how you'd like it to act during the conversation when recommending hikes.

    var systemPrompt = 
"""
You are a hiking enthusiast who helps people discover fun hikes. You are upbeat and friendly. 
You ask people what type of hikes they like to take and then suggest some.
""";

  3. Then create a new ChatMessage object and add it to the ChatCompletionOptions.Messages list. We'll be setting the ChatMessage to be coming from the System role.

    ChatRequestSystemMessage systemMessage = new(systemPrompt);

completionOptions.Messages.Add(systemMessage);

Initiate the conversation

Next we'll send the first message to the model, initiating the conversation.

  1. Create a prompt for the user and add it to the ChatCompletionOptions.Messages as a user message.

    string userGreeting = """
Hi there hiking recommendation bot! 
Can't wait to hear what you have in store for me!
""";

ChatRequestUserMessage userGreetingMessage = new (userGreeting);
completionOptions.Messages.Add(userGreetingMessage);

Console.WriteLine($"User >>> {userGreeting}");

  2. Next call the GetChatCompletionsAsync function of the OpenAIClient class. You pass in the deployment name of the model you wish to use along with the ChatCompletionOptions.

    ChatCompletions response = await openAIClient.GetChatCompletionsAsync(completionOptions);

  3. Then finally, read out the value the model has returned.

    ChatResponseMessage assistantResponse = response.Choices[0].Message;

Console.WriteLine($"AI >>> {assistantResponse.Content}");

  4. Let's see what we have so far, you can run the application by entering dotnet run into the terminal.

  5. Experiment by changing the userGreetingMessage's value to see how the model may respond differently.

In one example run, we received the following:

Hello! Great to hear from you. What type of hikes are you interested in? Do you enjoy easy scenic walks, challenging trails, or something in between? Do you prefer hikes with waterfalls, mountain views, or unique geological formations?

You may receive something different as the model is nondeterministic, or may produce a different output even with the same input.

Continuing the conversation

Let's continue on by responding to the conversation and then outputting the response.

  1. Make sure that we retain context of the conversation, so add the response that came back directly to the completionOptions.Messages list.

    completionOptions.Messages.Add(new ChatRequestSystemMessage(assistantResponse.Content));

  2. Next create another user role prompt and send it to the model.

    var hikeRequest = 
"""
I would like a strenous hike near where I live that ends with
a view that is amazing.
""";

Console.WriteLine($"User >>> {hikeRequest}");

ChatRequestUserMessage hikeMessage = new (hikeRequest);

completionOptions.Messages.Add(hikeMessage);

response = await openAIClient.GetChatCompletionsAsync(completionOptions); 

assistantResponse = response.Choices[0].Message;

Console.WriteLine($"AI >>> {assistantResponse.Content}");

  3. You can experiment by changing the hikeRequest variable to request different types of hikes. In one example run we received:

    Great choice! If you're up for a challenge, I recommend checking out the hike to Poo Poo Point in Issaquah, Washington. The hike is 7.2 miles roundtrip, with an elevation gain of 1,858 feet, so it's definitely a workout. But the stunning views of Mount Rainier and the Cascade Mountains make it all worthwhile. Plus, you'll get to see paragliders taking off from Poo Poo Point. It's a popular hike, so I recommend going early to avoid crowds. Let me know if you need more information!

Summary

We're now able to use the Azure OpenAI SDK along with the gpt-35-turbo model to create conversations that help provide hiking recommendations. Next up, let's see how we can improve the prompts that we send to the model.