Function Calling 是賦予LLMs應用強大能力的一個很重要的機制，它可以使得LLM模型增強其延伸的能力(但不是LLMs真的自已能調用工具)。

假設我們為 Kernel 掛載一個可以取得天氣資訊的 Plugin

var builder = Kernel.CreateBuilder()
            .AddAzureOpenAIChatCompletion(
                endpoint: Config.aoai_endpoint,
                deploymentName: Config.aoai_deployment,
                apiKey: Config.aoai_apiKey);
builder.Plugins.AddFromType<WeatherPlugin>();

在 Semantic Kernel 裡實現 Function Calling 有二種方式

• 手動調用

var weather = await kernel.Plugins.GetFunction("WeatherPlugin", "GetWeather").InvokeAsync(kernel);

• 自動調用

OpenAIPromptExecutionSettings settings = new() { ToolCallBehavior = ToolCallBehavior.AutoInvokeKernelFunctions };

在手動調用的情況，我們可以控制何時調用，取得的結果是什麼，但在自動調用的情況下，何時被調用了？取得的結果內容是什麼？有時候我們可能需要監看或記錄它，以做為後續的改善調整，在 Semantic Kernel 裡提供了一個 IAutoFunctionInvocationFilter 介面可以讓開發者實作

• 實作 IAutoFunctionInvocationFilter
  取得被調用的Plugin Function相關資訊，包含 Function Name、Plugin Name、Function Result等。其中 await next(context) 會決定 pipeline 是否往下或是到此為止。有寫過ASP.MVC 或 ASP.NET Web API Filter的開發者面應該會滿眼熟的。

public class AutoFunctionInvocationFilter : IAutoFunctionInvocationFilter
{
    public async Task OnAutoFunctionInvocationAsync(AutoFunctionInvocationContext context, Func<AutoFunctionInvocationContext, Task> next)
    {
        //get function information
        Console.WriteLine($"functionName: {context.Function.Name}");
        Console.WriteLine($"PluginName: {context.Function.PluginName}");

        // Calling next filter in pipeline or function itself.
        // By skipping this call, next filters and function won't be invoked, and function call loop will proceed to the next function.
        await next(context);

        // Example: get function result
        Console.WriteLine($"function result: {Regex.Unescape(context.Result.ToString())}");

        // Example: override function result value
        // context.Result = new FunctionResult(context.Result, "Result from auto function invocation filter");

        // Example: Terminate function invocation
        //context.Terminate = false;
    }
}

• kernel 掛上 IAutoFunctionInvocationFilter 實作

builder.Services.AddSingleton<IAutoFunctionInvocationFilter>(new AutoFunctionInvocationFilter());

這樣便可以追蹤監看在 Auto Function Calling 的情況下，調用了哪些 Function，分別得到了什麼結果，甚至可以在 Filter 裡改寫 Function 原始執行後結果。

• 改寫FunctionResult

context.Result = new FunctionResult(context.Result, "Result from auto function invocation filter");

完整程式碼

public async Task Run()
{
    // use Azure OpenAI
    var builder = Kernel.CreateBuilder()
        .AddAzureOpenAIChatCompletion(
            endpoint: Config.aoai_endpoint,
            deploymentName: Config.aoai_deployment,
            apiKey: Config.aoai_apiKey);
    builder.Plugins.AddFromType<WeatherPlugin>();
    builder.Services.AddSingleton<IAutoFunctionInvocationFilter>(new AutoFunctionInvocationFilter());
    Kernel kernel = builder.Build();

    OpenAIPromptExecutionSettings settings = new() { ToolCallBehavior = ToolCallBehavior.AutoInvokeKernelFunctions };

    var prompt =
    """
    The weather is:\n {{WeatherPlugin.GetWeather}}
    想去高雄旅行，適合穿什麼衣服?
    """;
    Console.WriteLine(await kernel.InvokePromptAsync(prompt, new(settings)));
}

public class AutoFunctionInvocationFilter : IAutoFunctionInvocationFilter
{
    public async Task OnAutoFunctionInvocationAsync(AutoFunctionInvocationContext context, Func<AutoFunctionInvocationContext, Task> next)
    {
        //get function information
        Console.WriteLine($"functionName: {context.Function.Name}");
        Console.WriteLine($"PluginName: {context.Function.PluginName}");

        // Calling next filter in pipeline or function itself.
        // By skipping this call, next filters and function won't be invoked, and function call loop will proceed to the next function.
        await next(context);

        // Example: get function result
        Console.WriteLine($"function result: {Regex.Unescape(context.Result.ToString())}");

        // Example: override function result value
        // context.Result = new FunctionResult(context.Result, "Result from auto function invocation filter");

        // Example: Terminate function invocation
        //context.Terminate = false;
    }
}

• 執行結果

Hello, World GenAI !

functionName: GetWeather
PluginName: WeatherPlugin
function result: {"\u53F0\u5317":"\u60B6\u71B1, 35\u00B0C","\u53F0\u4E2D":"\u6674\u6717, 32\u00B0C","\u9AD8\u96C4":"\u708E\u71B1, 35\u00B0C","\u53F0\u5357":"\u6674\u6717, 34\u00B0C","\u53F0\u6771":"\u9670\u96E8, 28\u00B0C"}
function result: {"台北":"悶熱, 35°C","台中":"晴朗, 32°C","高雄":"炎熱, 35°C","台南":"晴朗, 34°C","台東":"陰雨, 28°C"}
在高雄的天氣是「晴朗，34°C」。這樣的高溫建議您穿著輕便、透氣的衣物，例如：

- 短袖T恤或襯衫
- 輕薄的短褲或裙子
- 遮陽帽和太陽眼鏡以防曬

記得多補充水分，避免在中午時分長時間暴露在陽光下！希望您在高雄有個愉快的旅程！