你当前正在访问 Microsoft Azure Global Edition 技术文档网站。 如果需要访问由世纪互联运营的 Microsoft Azure 中国技术文档网站,请访问 https://docs.azure.cn。

教程:使用 Foundry Local 生成多轮聊天助手

在本教程中,你将生成一个完全在设备上运行的交互式聊天助手。 助手在多个交换中维护对话上下文,因此它会记住你在对话前面讨论的内容。 使用 Foundry 本地 SDK 选择模型、定义系统提示,并逐个令牌流式输出响应。

本教程中,您将学习如何:

  • 设置项目并安装 Foundry 本地 SDK
  • 浏览模型目录并选择模型
  • 定义用于塑造助理行为的系统提示
  • 使用消息历史记录实现多轮次对话
  • 流式传输响应以提升响应体验
  • 在会话结束时清理资源

先决条件

  • 至少具有 8 GB RAM 的 Windows、macOS 或 Linux 计算机。

示例存储库

可以在 Foundry 示例GitHub存储库中找到本文的完整示例代码。 若要克隆存储库并导航到示例,请使用:

git clone https://github.com/microsoft-foundry/foundry-samples.git
cd foundry-samples/samples/csharp/foundry-local/tutorial-chat-assistant

安装软件包

dotnet add package Microsoft.AI.Foundry.Local
dotnet add package OpenAI

GitHub存储库中的 C# 示例是预配置项目。 如果要从头开始生成,则应阅读 Foundry Local SDK 参考 ,详细了解如何使用 Foundry Local 设置 C# 项目。

浏览目录并选择模型

Foundry Local SDK 提供了一个模型目录,其中列出了所有可用的模型。 在此步骤中,将初始化 SDK 并为聊天助手选择模型。

  • 打开 Program.cs 并将其内容替换为以下代码来初始化 SDK 并选择模型:

    CancellationToken ct = CancellationToken.None;
    
    var config = new Configuration
    {
        AppName = "foundry_local_samples",
        LogLevel = Microsoft.AI.Foundry.Local.LogLevel.Information
    };
    
    using var loggerFactory = LoggerFactory.Create(builder =>
    {
        builder.SetMinimumLevel(Microsoft.Extensions.Logging.LogLevel.Information);
    });
    var logger = loggerFactory.CreateLogger<Program>();
    
    // Initialize the singleton instance
    await FoundryLocalManager.CreateAsync(config, logger);
    var mgr = FoundryLocalManager.Instance;
    
    // Download and register all execution providers.
    var currentEp = "";
    await mgr.DownloadAndRegisterEpsAsync((epName, percent) =>
    {
        if (epName != currentEp)
        {
            if (currentEp != "") Console.WriteLine();
            currentEp = epName;
        }
        Console.Write($"\r  {epName.PadRight(30)}  {percent,6:F1}%");
    });
    if (currentEp != "") Console.WriteLine();
    
    // Select and load a model from the catalog
    var catalog = await mgr.GetCatalogAsync();
    var model = await catalog.GetModelAsync("qwen2.5-0.5b")
        ?? throw new Exception("Model not found");
    
    await model.DownloadAsync(progress =>
    {
        Console.Write($"\rDownloading model: {progress:F2}%");
        if (progress >= 100f) Console.WriteLine();
    });
    
    await model.LoadAsync();
    Console.WriteLine("Model loaded and ready.");
    
    // Create one stateful native chat session for the conversation.
    using var chatSession = new ChatSession(model);
    chatSession.SetStreaming(true);
    

    该方法 GetModelAsync 接受模型别名,该别名是映射到目录中特定模型的短友好名称。 该方法 DownloadAsync 将模型权重提取到本地缓存,并使 LoadAsync 模型准备好进行推理。

定义系统提示

系统提示设置助手的个性和行为。 它是对话历史记录中的第一条消息,模型在整个对话中引用它。

添加系统提示以调整助手的响应方式:

// Start the conversation with a system prompt
var systemPrompt = "You are a helpful, friendly assistant. Keep your responses " +
                   "concise and conversational. If you don't know something, say so.";

小窍门

尝试使用不同的系统提示来更改助手的行为。 例如,可以指示它作为海盗、教师或域专家做出响应。

实现多回合对话

聊天助手需要在多轮对话中保持上下文。 通过保留所有消息的列表(系统、用户和助理),并发送包含每个请求的完整列表来实现此目的。 模型使用此历史记录生成上下文相关的响应。

添加一个对话循环,该循环:

  • 从控制台读取用户输入。
  • 将用户消息追加到历史记录。
  • 将完整历史记录发送到模型。
  • 将助手的响应追加到历史记录中以供下一轮使用。
while (true)
{
    Console.Write("You: ");
    var userInput = Console.ReadLine();
    if (string.IsNullOrWhiteSpace(userInput) ||
        userInput.Equals("quit", StringComparison.OrdinalIgnoreCase) ||
        userInput.Equals("exit", StringComparison.OrdinalIgnoreCase))
    {
        break;
    }

    using var request = new Request();
    if (chatSession.TurnCount == 0)
    {
        request.AddItem(MessageItem.System(systemPrompt));
    }
    request.AddItem(MessageItem.User(userInput));

    // Stream the response token by token
    Console.Write("Assistant: ");
    await using var streamingResponse = chatSession.ProcessStreamingRequestAsync(request, ct);
    await foreach (var item in streamingResponse)
    {
        using (item)
        {
            if (item is TextItem { Type: TextItemType.Default } text)
            {
                Console.Write(text.Text);
                Console.Out.Flush();
            }
        }
    }
    using var finalResponse = await streamingResponse.FinalResponse;
    Console.WriteLine("\n");
}

每次调用CompleteChatAsync 都会接收完整的消息历史记录。 这就是模型“记住”之前轮次的方式 , 它不会在调用之间存储状态。

添加流响应

流式输出在生成每个令牌时进行打印,从而让助手显得更具响应性。 将 CompleteChatAsync 调用替换为 CompleteChatStreamingAsync 以逐个令牌流式传输响应。

更新对话循环以使用流式输出:

// Stream the response token by token
Console.Write("Assistant: ");
await using var streamingResponse = chatSession.ProcessStreamingRequestAsync(request, ct);
await foreach (var item in streamingResponse)
{
    using (item)
    {
        if (item is TextItem { Type: TextItemType.Default } text)
        {
            Console.Write(text.Text);
            Console.Out.Flush();
        }
    }
}
using var finalResponse = await streamingResponse.FinalResponse;
Console.WriteLine("\n");

流版本会累积完整响应,以便在流完成后将其添加到聊天历史记录中。

完整代码

将 Program.cs 的内容替换为完整的代码:

using Microsoft.AI.Foundry.Local;
using Microsoft.Extensions.Logging;

CancellationToken ct = CancellationToken.None;

var config = new Configuration
{
    AppName = "foundry_local_samples",
    LogLevel = Microsoft.AI.Foundry.Local.LogLevel.Information
};

using var loggerFactory = LoggerFactory.Create(builder =>
{
    builder.SetMinimumLevel(Microsoft.Extensions.Logging.LogLevel.Information);
});
var logger = loggerFactory.CreateLogger<Program>();

// Initialize the singleton instance
await FoundryLocalManager.CreateAsync(config, logger);
var mgr = FoundryLocalManager.Instance;

// Download and register all execution providers.
var currentEp = "";
await mgr.DownloadAndRegisterEpsAsync((epName, percent) =>
{
    if (epName != currentEp)
    {
        if (currentEp != "") Console.WriteLine();
        currentEp = epName;
    }
    Console.Write($"\r  {epName.PadRight(30)}  {percent,6:F1}%");
});
if (currentEp != "") Console.WriteLine();

// Select and load a model from the catalog
var catalog = await mgr.GetCatalogAsync();
var model = await catalog.GetModelAsync("qwen2.5-0.5b")
    ?? throw new Exception("Model not found");

await model.DownloadAsync(progress =>
{
    Console.Write($"\rDownloading model: {progress:F2}%");
    if (progress >= 100f) Console.WriteLine();
});

await model.LoadAsync();
Console.WriteLine("Model loaded and ready.");

// Create one stateful native chat session for the conversation.
using var chatSession = new ChatSession(model);
chatSession.SetStreaming(true);

// Start the conversation with a system prompt
var systemPrompt = "You are a helpful, friendly assistant. Keep your responses " +
                   "concise and conversational. If you don't know something, say so.";

Console.WriteLine("\nChat assistant ready! Type 'quit' to exit.\n");

while (true)
{
    Console.Write("You: ");
    var userInput = Console.ReadLine();
    if (string.IsNullOrWhiteSpace(userInput) ||
        userInput.Equals("quit", StringComparison.OrdinalIgnoreCase) ||
        userInput.Equals("exit", StringComparison.OrdinalIgnoreCase))
    {
        break;
    }

    using var request = new Request();
    if (chatSession.TurnCount == 0)
    {
        request.AddItem(MessageItem.System(systemPrompt));
    }
    request.AddItem(MessageItem.User(userInput));

    // Stream the response token by token
    Console.Write("Assistant: ");
    await using var streamingResponse = chatSession.ProcessStreamingRequestAsync(request, ct);
    await foreach (var item in streamingResponse)
    {
        using (item)
        {
            if (item is TextItem { Type: TextItemType.Default } text)
            {
                Console.Write(text.Text);
                Console.Out.Flush();
            }
        }
    }
    using var finalResponse = await streamingResponse.FinalResponse;
    Console.WriteLine("\n");
}

// Clean up - unload the model
chatSession.Dispose();
await model.UnloadAsync();
Console.WriteLine("Model unloaded. Goodbye!");

运行聊天助理:

dotnet run

会看到类似于以下内容的输出:

Downloading model: 100.00%
Model loaded and ready.

Chat assistant ready! Type 'quit' to exit.

You: What is photosynthesis?
Assistant: Photosynthesis is the process plants use to convert sunlight, water, and carbon
dioxide into glucose and oxygen. It mainly happens in the leaves, inside structures
called chloroplasts.

You: Why is it important for other living things?
Assistant: It's essential because photosynthesis produces the oxygen that most living things
breathe. It also forms the base of the food chain — animals eat plants or eat other
animals that depend on plants for energy.

You: quit
Model unloaded. Goodbye!

请注意助手如何记住先前轮次的上下文 — 当你问 “为什么对其他活物很重要?” 时, 它知道你仍在谈论光合作用。

示例存储库

本文的完整示例代码在 foundry-samples GitHub 存储库中提供。 若要克隆存储库并导航到示例,请使用:

git clone https://github.com/microsoft-foundry/foundry-samples.git
cd foundry-samples/samples/javascript/foundry-local/tutorial-chat-assistant

安装软件包

npm install foundry-local-sdk openai

浏览目录并选择模型

Foundry Local SDK 提供了一个模型目录,其中列出了所有可用的模型。 在此步骤中,将初始化 SDK 并为聊天助手选择模型。

  1. 创建名为 的文件 index.js。

  2. 添加以下代码以初始化 SDK 并选择模型:

    // Initialize the Foundry Local SDK
    const manager = FoundryLocalManager.create({
        appName: 'foundry_local_samples',
        logLevel: 'info'
    });
    
    // Download and register all execution providers.
    let currentEp = '';
    await manager.downloadAndRegisterEps((epName, percent) => {
        if (epName !== currentEp) {
            if (currentEp !== '') process.stdout.write('\n');
            currentEp = epName;
        }
        process.stdout.write(`\r  ${epName.padEnd(30)}  ${percent.toFixed(1).padStart(5)}%`);
    });
    if (currentEp !== '') process.stdout.write('\n');
    
    // Select and load a model from the catalog
    const model = await manager.catalog.getModel('qwen2.5-0.5b');
    
    await model.download((progress) => {
        process.stdout.write(`\rDownloading model: ${progress.toFixed(2)}%`);
    });
    console.log('\nModel downloaded.');
    
    await model.load();
    console.log('Model loaded and ready.');
    
    // Create a stateful native chat session
    const chatSession = new ChatSession(model);
    chatSession.setOptions({
        search: { temperature: 0.7, maxOutputTokens: 512 }
    });
    

    该方法 getModel 接受模型别名,该别名是映射到目录中特定模型的短友好名称。 该方法 download 将模型权重提取到本地缓存,并使 load 模型准备好进行推理。

定义系统提示

系统提示设置助手的个性和行为。 它是对话历史记录中的第一条消息,模型在整个对话中引用它。

添加系统提示以调整助手的响应方式:

// Start the conversation with a system prompt
const systemPrompt =
    'You are a helpful, friendly assistant. Keep your responses ' +
    'concise and conversational. If you don\'t know something, say so.';
let firstTurn = true;

小窍门

尝试使用不同的系统提示来更改助手的行为。 例如,可以指示它作为海盗、教师或域专家做出响应。

实现多回合对话

聊天助手需要在多轮对话中保持上下文。 通过保留所有消息的列表(系统、用户和助理),并发送包含每个请求的完整列表来实现此目的。 模型使用此历史记录生成上下文相关的响应。

添加一个对话循环,该循环:

  • 从控制台读取用户输入。
  • 将用户消息追加到历史记录。
  • 将完整历史记录发送到模型。
  • 将助手的响应追加到历史记录中以供下一轮使用。
while (true) {
    const userInput = await askQuestion('You: ');
    if (userInput.trim().toLowerCase() === 'quit' ||
        userInput.trim().toLowerCase() === 'exit') {
        break;
    }

    const request = new Request();
    if (firstTurn) {
        request.addItem(Item.systemMessage(systemPrompt));
        firstTurn = false;
    }
    request.addItem(Item.userMessage(userInput));

    // Stream the response token by token
    process.stdout.write('Assistant: ');
    const stream = chatSession.processStreamingRequest(request);
    for await (const item of stream) {
        if (item.type === 'text' && item.textType === 'default') {
            process.stdout.write(item.text);
        }
    }
    await stream.response;
    console.log('\n');
}

每次调用completeChat 都会接收完整的消息历史记录。 这就是模型“记住”之前轮次的方式 , 它不会在调用之间存储状态。

添加流响应

流式输出在生成每个令牌时进行打印,从而让助手显得更具响应性。 将 completeChat 调用替换为 completeStreamingChat 以逐个令牌流式传输响应。

更新对话循环以使用流式输出:

// Stream the response token by token
process.stdout.write('Assistant: ');
const stream = chatSession.processStreamingRequest(request);
for await (const item of stream) {
    if (item.type === 'text' && item.textType === 'default') {
        process.stdout.write(item.text);
    }
}
await stream.response;
console.log('\n');

流版本会累积完整响应,以便在流完成后将其添加到聊天历史记录中。

完整代码

创建一个名为 index.js 并添加以下完整代码的文件:

import {
    ChatSession,
    FoundryLocalManager,
    Item,
    Request
} from 'foundry-local-sdk';
import * as readline from 'readline';

// Initialize the Foundry Local SDK
const manager = FoundryLocalManager.create({
    appName: 'foundry_local_samples',
    logLevel: 'info'
});

// Download and register all execution providers.
let currentEp = '';
await manager.downloadAndRegisterEps((epName, percent) => {
    if (epName !== currentEp) {
        if (currentEp !== '') process.stdout.write('\n');
        currentEp = epName;
    }
    process.stdout.write(`\r  ${epName.padEnd(30)}  ${percent.toFixed(1).padStart(5)}%`);
});
if (currentEp !== '') process.stdout.write('\n');

// Select and load a model from the catalog
const model = await manager.catalog.getModel('qwen2.5-0.5b');

await model.download((progress) => {
    process.stdout.write(`\rDownloading model: ${progress.toFixed(2)}%`);
});
console.log('\nModel downloaded.');

await model.load();
console.log('Model loaded and ready.');

// Create a stateful native chat session
const chatSession = new ChatSession(model);
chatSession.setOptions({
    search: { temperature: 0.7, maxOutputTokens: 512 }
});

// Start the conversation with a system prompt
const systemPrompt =
    'You are a helpful, friendly assistant. Keep your responses ' +
    'concise and conversational. If you don\'t know something, say so.';
let firstTurn = true;

// Set up readline for console input
const rl = readline.createInterface({
    input: process.stdin,
    output: process.stdout
});

const askQuestion = (prompt) => new Promise((resolve) => {
    if (rl.closed) return resolve('quit');
    const onClose = () => resolve('quit');
    rl.once('close', onClose);
    try {
        rl.question(prompt, (answer) => {
            rl.off('close', onClose);
            resolve(answer);
        });
    } catch {
        rl.off('close', onClose);
        resolve('quit');
    }
});

console.log('\nChat assistant ready! Type \'quit\' to exit.\n');

while (true) {
    const userInput = await askQuestion('You: ');
    if (userInput.trim().toLowerCase() === 'quit' ||
        userInput.trim().toLowerCase() === 'exit') {
        break;
    }

    const request = new Request();
    if (firstTurn) {
        request.addItem(Item.systemMessage(systemPrompt));
        firstTurn = false;
    }
    request.addItem(Item.userMessage(userInput));

    // Stream the response token by token
    process.stdout.write('Assistant: ');
    const stream = chatSession.processStreamingRequest(request);
    for await (const item of stream) {
        if (item.type === 'text' && item.textType === 'default') {
            process.stdout.write(item.text);
        }
    }
    await stream.response;
    console.log('\n');
}

// Clean up - unload the model
chatSession.dispose();
await model.unload();
console.log('Model unloaded. Goodbye!');
rl.close();

运行聊天助理:

node index.js

会看到类似于以下内容的输出:

Downloading model: 100.00%
Model downloaded.
Model loaded and ready.

Chat assistant ready! Type 'quit' to exit.

You: What is photosynthesis?
Assistant: Photosynthesis is the process plants use to convert sunlight, water, and carbon
dioxide into glucose and oxygen. It mainly happens in the leaves, inside structures
called chloroplasts.

You: Why is it important for other living things?
Assistant: It's essential because photosynthesis produces the oxygen that most living things
breathe. It also forms the base of the food chain — animals eat plants or eat other
animals that depend on plants for energy.

You: quit
Model unloaded. Goodbye!

请注意助手如何记住先前轮次的上下文 — 当你问 “为什么对其他活物很重要?” 时, 它知道你仍在谈论光合作用。

示例存储库

本文的完整示例代码在 foundry-samples GitHub 存储库中提供。 若要克隆存储库并导航到示例,请使用:

git clone https://github.com/microsoft-foundry/foundry-samples.git
cd foundry-samples/samples/python/foundry-local/tutorial-chat-assistant

安装软件包

pip install foundry-local-sdk openai

浏览目录并选择模型

Foundry Local SDK 提供了一个模型目录,其中列出了所有可用的模型。 在此步骤中,将初始化 SDK 并为聊天助手选择模型。

  1. 创建名为 的文件 main.py。

  2. 添加以下代码以初始化 SDK 并选择模型:

    # Initialize the Foundry Local SDK
    config = Configuration(app_name="foundry_local_samples")
    FoundryLocalManager.initialize(config)
    manager = FoundryLocalManager.instance
    
    # Discover and register execution providers that are not already available.
    missing_eps = [ep.name for ep in manager.discover_eps() if not ep.is_registered]
    if missing_eps:
        manager.download_and_register_eps(missing_eps)
    
    # Select and load a model from the catalog.
    model = manager.catalog.get_model("qwen2.5-0.5b")
    if model is None:
        raise RuntimeError("Model 'qwen2.5-0.5b' was not found in the catalog.")
    if not model.is_cached:
        model.download(
            lambda progress: print(
                f"\rDownloading model: {progress:.2f}%", end="", flush=True
            )
        )
        print()
    model.load()
    print("Model loaded and ready.")
    

    该方法 get_model 接受模型别名,该别名是映射到目录中特定模型的短友好名称。 该方法 download 将模型权重提取到本地缓存,并使 load 模型准备好进行推理。

定义系统提示

系统提示设置助手的个性和行为。 它是对话历史记录中的第一条消息,模型在整个对话中引用它。

添加系统提示以调整助手的响应方式:

# Start the conversation with a system prompt.
system_prompt = (
    "You are a helpful, friendly assistant. Keep your responses "
    "concise and conversational. If you don't know something, say so."
)

小窍门

尝试使用不同的系统提示来更改助手的行为。 例如,可以指示它作为海盗、教师或域专家做出响应。

实现多回合对话

聊天助手需要在多轮对话中保持上下文。 通过保留所有消息的列表(系统、用户和助理),并发送包含每个请求的完整列表来实现此目的。 模型使用此历史记录生成上下文相关的响应。

添加一个对话循环,该循环:

  • 从控制台读取用户输入。
  • 将用户消息追加到历史记录。
  • 将完整历史记录发送到模型。
  • 将助手的响应追加到历史记录中以供下一轮使用。
while True:
    user_input = input("You: ")
    if user_input.strip().lower() in ("quit", "exit"):
        break

    # The stateful session retains earlier turns automatically.
    with Request() as request:
        if first_turn:
            request.add_item(MessageItem.system(system_prompt))
            first_turn = False
        request.add_item(MessageItem.user(user_input))
        print("Assistant: ", end="", flush=True)
        with session.process_streaming_request(request) as stream:
            for item in stream:
                if (
                    isinstance(item, TextItem)
                    and item.type == TextItemType.DEFAULT
                ):
                    print(item.text, end="", flush=True)
        print("\n")

每次调用complete_chat 都会接收完整的消息历史记录。 这就是模型“记住”之前轮次的方式 , 它不会在调用之间存储状态。

添加流响应

流式输出在生成每个令牌时进行打印,从而让助手显得更具响应性。 将 complete_chat 调用替换为 complete_streaming_chat 以逐个令牌流式传输响应。

更新对话循环以使用流式输出:

# The stateful session retains earlier turns automatically.
with Request() as request:
    if first_turn:
        request.add_item(MessageItem.system(system_prompt))
        first_turn = False
    request.add_item(MessageItem.user(user_input))
    print("Assistant: ", end="", flush=True)
    with session.process_streaming_request(request) as stream:
        for item in stream:
            if (
                isinstance(item, TextItem)
                and item.type == TextItemType.DEFAULT
            ):
                print(item.text, end="", flush=True)
    print("\n")

流版本会累积完整响应,以便在流完成后将其添加到聊天历史记录中。

完整代码

创建一个名为 main.py 并添加以下完整代码的文件:

from foundry_local_sdk import (
    ChatSession,
    Configuration,
    FoundryLocalManager,
    MessageItem,
    Request,
    RequestOptions,
    SearchOptions,
    TextItem,
    TextItemType,
)



def main():
    # Initialize the Foundry Local SDK
    config = Configuration(app_name="foundry_local_samples")
    FoundryLocalManager.initialize(config)
    manager = FoundryLocalManager.instance

    # Discover and register execution providers that are not already available.
    missing_eps = [ep.name for ep in manager.discover_eps() if not ep.is_registered]
    if missing_eps:
        manager.download_and_register_eps(missing_eps)

    # Select and load a model from the catalog.
    model = manager.catalog.get_model("qwen2.5-0.5b")
    if model is None:
        raise RuntimeError("Model 'qwen2.5-0.5b' was not found in the catalog.")
    if not model.is_cached:
        model.download(
            lambda progress: print(
                f"\rDownloading model: {progress:.2f}%", end="", flush=True
            )
        )
        print()
    model.load()
    print("Model loaded and ready.")

    try:
        # Start the conversation with a system prompt.
        system_prompt = (
            "You are a helpful, friendly assistant. Keep your responses "
            "concise and conversational. If you don't know something, say so."
        )

        print("\nChat assistant ready! Type 'quit' to exit.\n")

        with ChatSession(model) as session:
            session.set_options(
                RequestOptions(
                    search=SearchOptions(temperature=0.7, max_output_tokens=512)
                )
            )
            session.set_streaming(True)
            first_turn = True

            while True:
                user_input = input("You: ")
                if user_input.strip().lower() in ("quit", "exit"):
                    break

                # The stateful session retains earlier turns automatically.
                with Request() as request:
                    if first_turn:
                        request.add_item(MessageItem.system(system_prompt))
                        first_turn = False
                    request.add_item(MessageItem.user(user_input))
                    print("Assistant: ", end="", flush=True)
                    with session.process_streaming_request(request) as stream:
                        for item in stream:
                            if (
                                isinstance(item, TextItem)
                                and item.type == TextItemType.DEFAULT
                            ):
                                print(item.text, end="", flush=True)
                    print("\n")
    finally:
        # Clean up - unload the model
        if model.is_loaded:
            model.unload()
        manager.close()
        print("Model unloaded. Goodbye!")


if __name__ == "__main__":
    main()

运行聊天助理:

python main.py

会看到类似于以下内容的输出:

Downloading model: 100.00%
Model loaded and ready.

Chat assistant ready! Type 'quit' to exit.

You: What is photosynthesis?
Assistant: Photosynthesis is the process plants use to convert sunlight, water, and carbon
dioxide into glucose and oxygen. It mainly happens in the leaves, inside structures
called chloroplasts.

You: Why is it important for other living things?
Assistant: It's essential because photosynthesis produces the oxygen that most living things
breathe. It also forms the base of the food chain — animals eat plants or eat other
animals that depend on plants for energy.

You: quit
Model unloaded. Goodbye!

请注意助手如何记住先前轮次的上下文 — 当你问 “为什么对其他活物很重要?” 时, 它知道你仍在谈论光合作用。

示例存储库

本文的完整示例代码在 foundry-samples GitHub 存储库中提供。 若要克隆存储库并导航到示例,请使用:

git clone https://github.com/microsoft-foundry/foundry-samples.git
cd foundry-samples/samples/rust/foundry-local/tutorial-chat-assistant

安装软件包

cargo add foundry-local-sdk
cargo add tokio --features full
cargo add tokio-stream anyhow

浏览目录并选择模型

Foundry Local SDK 提供了一个模型目录,其中列出了所有可用的模型。 在此步骤中,将初始化 SDK 并为聊天助手选择模型。

  • 打开 src/main.rs 并将其内容替换为以下代码来初始化 SDK 并选择模型:

    // Initialize the Foundry Local SDK
    let manager = FoundryLocalManager::create(FoundryLocalConfig::new("foundry_local_samples"))?;
    
    manager
        .download_and_register_eps_with_progress(None, {
            let mut current_ep = String::new();
            move |ep_name: &str, percent: f64| {
                if ep_name != current_ep {
                    if !current_ep.is_empty() {
                        println!();
                    }
                    current_ep = ep_name.to_string();
                }
                print!("\r  {ep_name:<30}  {percent:5.1}%");
                io::stdout().flush().ok();
            }
        })
        .await?;
    println!();
    
    let model = manager.catalog().get_model("qwen2.5-0.5b").await?;
    if !model.is_cached().await? {
        println!("Downloading model...");
        model
            .download(Some(|progress: f64| {
                print!("\r  {progress:.1}%");
                io::stdout().flush().ok();
            }))
            .await?;
        println!();
    }
    
    model.load().await?;
    let session = ChatSession::new(&model).await?;
    session
        .set_options(RequestOptions {
            search: SearchOptions {
                temperature: Some(0.7),
                max_output_tokens: Some(512),
                ..Default::default()
            },
            ..Default::default()
        })
        .await?;
    println!("Model loaded and ready.");
    

    该方法 get_model 接受模型别名,该别名是映射到目录中特定模型的短友好名称。 该方法 download 将模型权重提取到本地缓存,并使 load 模型准备好进行推理。

定义系统提示

系统提示设置助手的个性和行为。 它是对话历史记录中的第一条消息,模型在整个对话中引用它。

添加系统提示以调整助手的响应方式:

let system_prompt = "You are a helpful, friendly assistant. Keep your responses concise and \
                     conversational. If you don't know something, say so.";

小窍门

尝试使用不同的系统提示来更改助手的行为。 例如,可以指示它作为海盗、教师或域专家做出响应。

实现多回合对话

聊天助手需要在多轮对话中保持上下文。 通过保留所有消息(系统、用户和助理)的向量并发送包含每个请求的完整列表来实现此目的。 模型使用此历史记录生成上下文相关的响应。

添加一个对话循环,该循环:

  • 从控制台读取用户输入。
  • 将用户消息追加到历史记录。
  • 将完整历史记录发送到模型。
  • 将助手的响应追加到历史记录中以供下一轮使用。
loop {
    print!("You: ");
    io::stdout().flush()?;

    let mut input = String::new();
    stdin.lock().read_line(&mut input)?;
    let input = input.trim();

    if input.eq_ignore_ascii_case("quit") || input.eq_ignore_ascii_case("exit") {
        break;
    }
    if input.is_empty() {
        continue;
    }

    let mut items = Vec::new();
    if session.turn_count() == 0 {
        items.push(Item::system_message(vec![Item::text(system_prompt)]));
    }
    items.push(Item::user_message(vec![Item::text(input)]));

    print!("Assistant: ");
    io::stdout().flush()?;
    let mut stream = session.process_streaming_request(Request::from_items(items));
    while let Some(item) = stream.next().await {
        if let Item::Text {
            text,
            kind: TextKind::Default,
        } = item?
        {
            print!("{text}");
            io::stdout().flush()?;
        }
    }
    println!("\n");
}

每次调用complete_chat 都会接收完整的消息历史记录。 这就是模型“记住”之前轮次的方式 , 它不会在调用之间存储状态。

添加流响应

流式输出在生成每个令牌时进行打印,从而让助手显得更具响应性。 将 complete_chat 调用替换为 complete_streaming_chat 以逐个令牌流式传输响应。

更新对话循环以使用流式输出:

print!("Assistant: ");
io::stdout().flush()?;
let mut stream = session.process_streaming_request(Request::from_items(items));
while let Some(item) = stream.next().await {
    if let Item::Text {
        text,
        kind: TextKind::Default,
    } = item?
    {
        print!("{text}");
        io::stdout().flush()?;
    }
}
println!("\n");

流版本会累积完整响应,以便在流完成后将其添加到聊天历史记录中。

完整代码

将 src/main.rs 的内容替换为完整的代码:

use foundry_local_sdk::{
    ChatSession, FoundryLocalConfig, FoundryLocalManager, Item, Request, RequestOptions,
    SearchOptions, TextKind,
};
use std::io::{self, BufRead, Write};
use tokio_stream::StreamExt;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Initialize the Foundry Local SDK
    let manager = FoundryLocalManager::create(FoundryLocalConfig::new("foundry_local_samples"))?;

    manager
        .download_and_register_eps_with_progress(None, {
            let mut current_ep = String::new();
            move |ep_name: &str, percent: f64| {
                if ep_name != current_ep {
                    if !current_ep.is_empty() {
                        println!();
                    }
                    current_ep = ep_name.to_string();
                }
                print!("\r  {ep_name:<30}  {percent:5.1}%");
                io::stdout().flush().ok();
            }
        })
        .await?;
    println!();

    let model = manager.catalog().get_model("qwen2.5-0.5b").await?;
    if !model.is_cached().await? {
        println!("Downloading model...");
        model
            .download(Some(|progress: f64| {
                print!("\r  {progress:.1}%");
                io::stdout().flush().ok();
            }))
            .await?;
        println!();
    }

    model.load().await?;
    let session = ChatSession::new(&model).await?;
    session
        .set_options(RequestOptions {
            search: SearchOptions {
                temperature: Some(0.7),
                max_output_tokens: Some(512),
                ..Default::default()
            },
            ..Default::default()
        })
        .await?;
    println!("Model loaded and ready.");

    let system_prompt = "You are a helpful, friendly assistant. Keep your responses concise and \
                         conversational. If you don't know something, say so.";

    println!("\nChat assistant ready! Type 'quit' to exit.\n");

    let stdin = io::stdin();
    loop {
        print!("You: ");
        io::stdout().flush()?;

        let mut input = String::new();
        stdin.lock().read_line(&mut input)?;
        let input = input.trim();

        if input.eq_ignore_ascii_case("quit") || input.eq_ignore_ascii_case("exit") {
            break;
        }
        if input.is_empty() {
            continue;
        }

        let mut items = Vec::new();
        if session.turn_count() == 0 {
            items.push(Item::system_message(vec![Item::text(system_prompt)]));
        }
        items.push(Item::user_message(vec![Item::text(input)]));

        print!("Assistant: ");
        io::stdout().flush()?;
        let mut stream = session.process_streaming_request(Request::from_items(items));
        while let Some(item) = stream.next().await {
            if let Item::Text {
                text,
                kind: TextKind::Default,
            } = item?
            {
                print!("{text}");
                io::stdout().flush()?;
            }
        }
        println!("\n");
    }

    drop(session);
    model.unload().await?;
    println!("Model unloaded. Goodbye!");

    Ok(())
}

运行聊天助理:

cargo run

会看到类似于以下内容的输出:

Downloading model: 100.00%
Model loaded and ready.

Chat assistant ready! Type 'quit' to exit.

You: What is photosynthesis?
Assistant: Photosynthesis is the process plants use to convert sunlight, water, and carbon
dioxide into glucose and oxygen. It mainly happens in the leaves, inside structures
called chloroplasts.

You: Why is it important for other living things?
Assistant: It's essential because photosynthesis produces the oxygen that most living things
breathe. It also forms the base of the food chain — animals eat plants or eat other
animals that depend on plants for energy.

You: quit
Model unloaded. Goodbye!

请注意助手如何记住先前轮次的上下文 — 当你问 “为什么对其他活物很重要?” 时, 它知道你仍在谈论光合作用。

示例存储库

克隆 Foundry 示例GitHub存储库并转到 C++ 示例:

git clone https://github.com/microsoft-foundry/foundry-samples.git
cd foundry-samples/samples/cpp/foundry-local/tutorial-chat-assistant

安装软件包

C++ 示例需要:

  • CMake 3.20 或更高版本。
  • 一个 C++20 编译器。
  • PATH 上的 NuGet CLI。

配置并生成示例:

cmake -S . -B build
cmake --build build --config Release

在配置期间,示例将版本 2.2.1 的 Microsoft.AI.Foundry.Local.Runtime NuGet 包还原到 build/_packages。 该包包含用于 Windows、Linux 和 macOS 的 C++ 标头和本机库。 不需要单独的 WinML 包。

CMake 配置会根据您的操作系统和架构选择运行时资源,并将所需的本机库复制到可执行文件旁边。 在 macOS 和 Linux 上,从 build中运行可执行文件。 在 Windows 上,运行位于 .exe 的 build 文件。

浏览目录并选择模型

初始化 SDK,从目录中选择模型,然后下载并加载它:

Configuration config("foundry_local_cpp_tutorial_chat");
Manager manager(std::move(config));
auto model = SelectChatModel(manager.GetCatalog());
if (!model) {
  throw std::runtime_error("No supported CPU chat model was found.");
}
if (!model->IsCached()) {
  model->Download([](float progress) {
    std::cout << "\rDownloading model: " << static_cast<int>(progress) << "%" << std::flush;
    return 0;
  });
  std::cout << "\n";
}
model->Load();

定义系统提示

添加系统提示以调整助手的响应方式:

Request system_prompt{
    SystemMessage(
        "You are a concise, friendly local assistant. State when you are unsure."),
};
session.ProcessRequest(system_prompt);

实现多回合对话

维护会话历史记录,并随每个请求一起发送它:

std::cout << "Enter a message, or type 'exit'.\n";
for (std::string input; std::cout << "You: " && std::getline(std::cin, input);) {
  if (input == "exit" || input == "quit") {
    break;
  }
  if (input.empty()) {
    continue;
  }

  EnableStreaming(session);
  Request request{UserMessage(input)};
  std::cout << "Assistant: ";
  session.ProcessRequest(request);
  std::cout << "\n";
}

添加流响应

在模型生成响应文本时流式传输响应文本:

void EnableStreaming(ChatSession& session) {
  session.SetStreamingCallback([](Item item) {
    if (item.GetType() == FOUNDRY_LOCAL_ITEM_TEXT &&
        item.GetText().type == FOUNDRY_LOCAL_TEXT_ITEM_TYPE_DEFAULT) {
      std::cout << item.GetText().text << std::flush;
    }
    return 0;
  });
}

完整代码

#include <foundry_local/foundry_local_cpp.h>

#include <iostream>
#include <memory>
#include <string>

using namespace foundry_local;

std::unique_ptr<IModel> SelectChatModel(ICatalog& catalog) {
  for (const std::string alias : {"qwen2.5-0.5b", "qwen3.5-0.8b"}) {
    auto model = catalog.GetModel(alias);
    if (!model) {
      continue;
    }
    for (const auto& variant : model->GetVariants()) {
      const auto info = variant->GetInfo();
      if (info.DeviceType() == FOUNDRY_LOCAL_DEVICE_CPU &&
          (info.Task() == "chat-completion" || info.Task() == "coding")) {
        model->SelectVariant(*variant);
        return model;
      }
    }
  }
  return nullptr;
}

void EnableStreaming(ChatSession& session);

int main() {
  try {
    Configuration config("foundry_local_cpp_tutorial_chat");
    Manager manager(std::move(config));
    auto model = SelectChatModel(manager.GetCatalog());
    if (!model) {
      throw std::runtime_error("No supported CPU chat model was found.");
    }
    if (!model->IsCached()) {
      model->Download([](float progress) {
        std::cout << "\rDownloading model: " << static_cast<int>(progress) << "%" << std::flush;
        return 0;
      });
      std::cout << "\n";
    }
    model->Load();

    {
      ChatSession session(*model);
      session.SetOptions({
          .search = {.temperature = 0.2F, .max_output_tokens = 512, .seed = 42},
      });

      Request system_prompt{
          SystemMessage(
              "You are a concise, friendly local assistant. State when you are unsure."),
      };
      session.ProcessRequest(system_prompt);

      std::cout << "Enter a message, or type 'exit'.\n";
      for (std::string input; std::cout << "You: " && std::getline(std::cin, input);) {
        if (input == "exit" || input == "quit") {
          break;
        }
        if (input.empty()) {
          continue;
        }

        EnableStreaming(session);
        Request request{UserMessage(input)};
        std::cout << "Assistant: ";
        session.ProcessRequest(request);
        std::cout << "\n";
      }
    }

    model->Unload();
    manager.Shutdown();
  } catch (const Error& error) {
    std::cerr << "Foundry Local error [" << error.Code() << "]: " << error.what() << "\n";
    return 1;
  } catch (const std::exception& error) {
    std::cerr << "Error: " << error.what() << "\n";
    return 1;
  }

  return 0;
}

void EnableStreaming(ChatSession& session) {
  session.SetStreamingCallback([](Item item) {
    if (item.GetType() == FOUNDRY_LOCAL_ITEM_TEXT &&
        item.GetText().type == FOUNDRY_LOCAL_TEXT_ITEM_TYPE_DEFAULT) {
      std::cout << item.GetText().text << std::flush;
    }
    return 0;
  });
}

运行聊天助理:

./build/tutorial-chat-assistant

清理资源

卸载模型后,模型权重将保留在本地缓存中。 这意味着下次运行应用程序时,将跳过下载步骤,模型加载速度更快。 除非需要回收磁盘空间,否则不需要进行额外的清理。