通过

在 .NET 控制台应用中使用聊天补全配置

在本指南中，你将生成一个 AI 聊天应用程序，并使用从 Azure 应用程序配置动态加载的聊天补全配置来循环访问提示。

Azure 应用配置 GitHub 存储库中提供了完整的示例源代码。

先决条件

创建控制台应用

  1. 为项目新建一个文件夹。 在新文件夹中，运行以下命令以创建新的 .NET 控制台应用项目：

    dotnet new console

  2. 在项目中安装以下 NuGet 包：

    dotnet add package Microsoft.Extensions.Configuration.AzureAppConfiguration
dotnet add package Microsoft.Extensions.Configuration.Binder
dotnet add package Azure.Identity
dotnet add package Azure.AI.OpenAI

  3. 打开 Program.cs 文件，并在文件顶部添加以下命名空间：

    using Microsoft.Extensions.Configuration;
using Microsoft.Extensions.Configuration.AzureAppConfiguration;
using Azure.Identity;
using Azure.AI.OpenAI;
using OpenAI.Chat;

  4. 通过调用 AddAzureAppConfigurationProgram.cs 文件中的方法连接到应用配置存储。

    可以使用 Microsoft Entra ID（建议）或连接字符串连接到应用程序配置。 在此示例中，会使用 Microsoft Entra ID 和 DefaultAzureCredential 对应用程序配置存储区进行身份验证。 按照以下 说明 将 应用配置数据读取者 角色分配给由 DefaultAzureCredential表示的标识。 运行应用程序前，请务必留出足够的时间让权限生效。

    TokenCredential credential = new DefaultAzureCredential();
IConfigurationRefresher refresher = null;

// Load configuration from Azure App Configuration
IConfiguration configuration = new ConfigurationBuilder()
    .AddAzureAppConfiguration(options =>
    {
        Uri endpoint = new(Environment.GetEnvironmentVariable("AZURE_APPCONFIGURATION_ENDPOINT") ??
            throw new InvalidOperationException("The environment variable 'AZURE_APPCONFIGURATION_ENDPOINT' is not set or is empty."));
        options.Connect(endpoint, credential)
            // Load all keys that start with "ChatApp:" and have no label.
            .Select("ChatApp:*")
            // Reload configuration if any selected key-values have changed.
            // Use the default refresh interval of 30 seconds. It can be overridden via refreshOptions.SetRefreshInterval.
            .ConfigureRefresh(refreshOptions =>
            {
                refreshOptions.RegisterAll();
            });

        refresher = options.GetRefresher();
    })
    .Build();

  5. 创建连接到 Azure OpenAI 资源的实例 AzureOpenAIClient 。 可以使用 Microsoft Entra ID 或 API 密钥进行身份验证。

    若要使用 Microsoft Entra ID 访问 Azure OpenAI 资源，请使用 DefaultAzureCredential。 将“认知服务 OpenAI 用户”角色分配给 DefaultAzureCredential 所表示的标识。 有关详细步骤，请参阅 Azure OpenAI 服务指南的基于角色的访问控制 。 运行应用程序前，请务必留出足够的时间让权限生效。

    // Retrieve the OpenAI connection information from the configuration
Uri openaiEndpoint = new(configuration["ChatApp:AzureOpenAI:Endpoint"]);
string deploymentName = configuration["ChatApp:AzureOpenAI:DeploymentName"];

// Initialize the AzureOpenAIClient
AzureOpenAIClient azureClient = new(openaiEndpoint, credential);
// Create a chat client
ChatClient chatClient = azureClient.GetChatClient(deploymentName);

    若要使用 API 密钥访问 Azure OpenAI 资源，请添加以下代码：

    // Initialize the AzureOpenAIClient
var apiKey = configuration["ChatApp:AzureOpenAI:ApiKey"];
AzureOpenAIClient client = new(openAIEndpoint, new AzureKeyCredential(apiKey));

    如果密钥 ChatApp：AzureOpenAI：ApiKey 是应用配置中的 Key Vault 引用，请确保将以下代码片段添加到 AddAzureAppConfiguration 调用中，并向 应用授予对 Key Vault 的访问权限。

    options.ConfigureKeyVault(keyVaultOptions =>
{
    keyVaultOptions.SetCredential(credential);
});

  6. 在 ChatCompletionConfiguration 文件中定义类。

    internal class ChatCompletionConfiguration
{
    [ConfigurationKeyName("model")]
    public string? Model { get; set; }

    [ConfigurationKeyName("messages")]
    public List<Message>? Messages { get; set; }

    [ConfigurationKeyName("max_tokens")]
    public int MaxTokens { get; set; }

    [ConfigurationKeyName("temperature")]
    public float Temperature { get; set; }

    [ConfigurationKeyName("top_p")]
    public float TopP { get; set; }
}

internal class Message
{
    [ConfigurationKeyName("role")]
    public required string Role { get; set; }

    [ConfigurationKeyName("content")]
    public string? Content { get; set; }
}

  7. 更新 Program.cs 文件以添加帮助程序方法来 GetChatMessages 处理聊天消息：

    // Helper method to convert configuration messages to ChatMessage objects
static IEnumerable<ChatMessage> GetChatMessages(ChatCompletionConfiguration chatCompletionConfiguration)
{
    return chatCompletionConfiguration.Messages.Select<Message, ChatMessage>(message => message.Role switch
    {
        "system" => ChatMessage.CreateSystemMessage(message.Content),
        "user" => ChatMessage.CreateUserMessage(message.Content),
        "assistant" => ChatMessage.CreateAssistantMessage(message.Content),
        _ => throw new ArgumentException($"Unknown role: {message.Role}", nameof(message.Role))
    });
}

  8. 接下来，更新 Program.cs 文件中的现有代码以从 Azure 应用配置刷新配置，将最新的 AI 配置值应用于聊天完成设置，并从 AI 模型检索响应。

    // Initialize chat conversation
var chatConversation = new List<ChatMessage>();
Console.WriteLine("Chat started! What's on your mind?");

while (true)
{
    // Get user input
    Console.Write("You: ");
    string? userInput = Console.ReadLine();

    // Exit if user input is empty
    if (string.IsNullOrEmpty(userInput))
    {
        Console.WriteLine("Exiting chat. Goodbye!");
        break;
    }

    chatConversation.Add(ChatMessage.CreateUserMessage(userInput));

    // Refresh the configuration from Azure App Configuration
    await refresher.RefreshAsync();

    // Configure chat completion with AI configuration
    var chatCompletionConfiguration = configuration.GetSection("ChatApp:ChatCompletion").Get<ChatCompletionConfiguration>();

    var requestOptions = new ChatCompletionOptions()
    {
        MaxOutputTokenCount = chatCompletionConfiguration.MaxTokens,
        Temperature = chatCompletionConfiguration.Temperature,
        TopP = chatCompletionConfiguration.TopP
    };

    // Get latest system message from AI configuration
    var chatMessages = new List<ChatMessage>(GetChatMessages(chatCompletionConfiguration));
    chatMessages.AddRange(chatConversation);

    // Get AI response and add it to chat conversation
    var response = await chatClient.CompleteChatAsync(chatMessages, requestOptions);
    string aiResponse = response.Value.Content[0].Text;

    Console.WriteLine($"AI: {aiResponse}");
    chatConversation.Add(ChatMessage.CreateAssistantMessage(aiResponse));

    Console.WriteLine();
}

  9. 完成前面的步骤后， Program.cs 文件现在应包含完整的实现，如下所示：

    using Azure.AI.OpenAI;
using Azure.Identity;
using Azure.Core;
using Microsoft.Extensions.Configuration;
using Microsoft.Extensions.Configuration.AzureAppConfiguration;
using OpenAI.Chat;

TokenCredential credential = new DefaultAzureCredential();
IConfigurationRefresher refresher = null;

// Load configuration from Azure App Configuration
IConfiguration configuration = new ConfigurationBuilder()
    .AddAzureAppConfiguration(options =>
    {
        Uri endpoint = new(Environment.GetEnvironmentVariable("AZURE_APPCONFIGURATION_ENDPOINT") ??
            throw new InvalidOperationException("The environment variable 'AZURE_APPCONFIGURATION_ENDPOINT' is not set or is empty."));
        options.Connect(endpoint, credential)
            // Load all keys that start with "ChatApp:" and have no label.
            .Select("ChatApp:*")
            // Reload configuration if any selected key-values have changed.
            // Use the default refresh interval of 30 seconds. It can be overridden via refreshOptions.SetRefreshInterval.
            .ConfigureRefresh(refreshOptions =>
            {
                refreshOptions.RegisterAll();
            });

        refresher = options.GetRefresher();
    })
    .Build();

// Retrieve the OpenAI connection information from the configuration
Uri openaiEndpoint = new(configuration["ChatApp:AzureOpenAI:Endpoint"]);
string deploymentName = configuration["ChatApp:AzureOpenAI:DeploymentName"];

// Create a chat client
AzureOpenAIClient azureClient = new(openaiEndpoint, credential);
ChatClient chatClient = azureClient.GetChatClient(deploymentName);

// Initialize chat conversation
var chatConversation = new List<ChatMessage>();
Console.WriteLine("Chat started! What's on your mind?");

while (true)
{
    // Get user input
    Console.Write("You: ");
    string? userInput = Console.ReadLine();

    // Exit if user input is empty
    if (string.IsNullOrEmpty(userInput))
    {
        Console.WriteLine("Exiting chat. Goodbye!");
        break;
    }

    chatConversation.Add(ChatMessage.CreateUserMessage(userInput));

    // Refresh the configuration from Azure App Configuration
    await refresher.RefreshAsync();

    // Configure chat completion with AI configuration
    var chatCompletionConfiguration = configuration.GetSection("ChatApp:ChatCompletion").Get<ChatCompletionConfiguration>();
    var requestOptions = new ChatCompletionOptions()
    {
        MaxOutputTokenCount = chatCompletionConfiguration.MaxTokens,
        Temperature = chatCompletionConfiguration.Temperature,
        TopP = chatCompletionConfiguration.TopP
    };

    // Get latest system message from AI configuration
    var chatMessages = new List<ChatMessage>(GetChatMessages(chatCompletionConfiguration));
    chatMessages.AddRange(chatConversation);

    // Get AI response and add it to chat conversation
    var response = await chatClient.CompleteChatAsync(chatMessages, requestOptions);
    string aiResponse = response.Value.Content[0].Text;

    Console.WriteLine($"AI: {aiResponse}");
    chatConversation.Add(ChatMessage.CreateAssistantMessage(aiResponse));

    Console.WriteLine();
}

// Helper method to convert configuration messages to ChatMessage objects
static IEnumerable<ChatMessage> GetChatMessages(ChatCompletionConfiguration chatCompletionConfiguration)
{
    return chatCompletionConfiguration.Messages.Select<Message, ChatMessage>(message => message.Role switch
    {
        "system" => ChatMessage.CreateSystemMessage(message.Content),
        "user" => ChatMessage.CreateUserMessage(message.Content),
        "assistant" => ChatMessage.CreateAssistantMessage(message.Content),
        _ => throw new ArgumentException($"Unknown role: {message.Role}", nameof(message.Role))
    });
}

internal class ChatCompletionConfiguration
{
    [ConfigurationKeyName("model")]
    public string? Model { get; set; }

    [ConfigurationKeyName("messages")]
    public List<Message>? Messages { get; set; }

    [ConfigurationKeyName("max_tokens")]
    public int MaxTokens { get; set; }

    [ConfigurationKeyName("temperature")]
    public float Temperature { get; set; }

    [ConfigurationKeyName("top_p")]
    public float TopP { get; set; }
}

internal class Message
{
    [ConfigurationKeyName("role")]
    public required string Role { get; set; }

    [ConfigurationKeyName("content")]
    public string? Content { get; set; }
}

生成并运行应用

  1. 将名为 “AZURE_APPCONFIGURATION_ENDPOINT ”的环境变量设置为 Azure 门户中应用商店 “概述 ”下找到的应用配置存储的终结点。

    如果使用 Windows 命令提示符，则请运行以下命令并重启命令提示符，这样更改才会生效：

    setx AZURE_APPCONFIGURATION_ENDPOINT "<endpoint-of-your-app-configuration-store>"

    如果使用 PowerShell，请运行以下命令：

    $Env:AZURE_APPCONFIGURATION_ENDPOINT = "<endpoint-of-your-app-configuration-store>"

    如果使用 macOS 或 Linux，请运行以下命令：

    export AZURE_APPCONFIGURATION_ENDPOINT ='<endpoint-of-your-app-configuration-store>'

  2. 正确设置环境变量后，运行以下命令以生成并运行应用。

    dotnet run

  3. 当系统提示输入“You：”时，键入消息“你的姓名是什么？”，然后按 Enter 键。

    Chat started! What's on your mind?
You: What is your name ?
AI: I'm your helpful assistant! I don't have a personal name, but you can call me whatever you'd like. 
😊 Do you have a name in mind?

  4. 在 Azure 门户中，选择已创建的应用配置存储实例。 从操作菜单中选择配置资源管理器，然后选择ChatApp:ChatCompletion键。 将 Messages 属性的值进行更新：

    • 角色：系统
    • 内容：“你是海盗，你的名字是埃迪。

  5. 出现“You：”提示时，键入相同的消息。 请务必等待片刻，刷新间隔过后，然后按 Enter 键查看输出中更新的 AI 响应。

    Chat started! What's on your mind?
You: What is your name ?
AI: I'm your helpful assistant! I don't have a personal name, but you can call me whatever you'd like. 
😊 Do you have a name in mind?

You: What is your name ?
AI: Arrr, matey! Me name be Eddy, the most fearsome pirate to ever sail the seven seas!
What be yer name, landlubber?