解锁GraphRag.Net的无限可能:手把手教你集成国产模型和本地模型

        在上次的文章中,我们已经详细介绍了GraphRag的基本功能和使用方式。如果你还不熟悉,建议先阅读前面的文章

        通过前两篇文章,相信你已经了解到GraphRag.Net目前只支持OpenAI规范的接口,但许多小伙伴在社区中提议,希望能增加对本地模型(例如:ollama等)的支持。所以这次,我们将探讨如何在GraphRag.Net中使用自定义模型和本地模型。

 

为什么选择GraphRag.Net?

        GraphRag.Net采用了Semantic Kernel作为基础,让我们能够非常简洁地抽象出会话与向量接口。因此,用户可以非常方便地实现自己定制的解决方案。接下来,我们会通过一个具体的例子,展示如何将本地模型和国产模型集成到GraphRag.Net中。

默认配置方法

        首先,我们来看看如何进行默认配置:

1
2
3
4
5
6
7
8
9
// OpenAI配置
builder.Configuration.GetSection("OpenAI").Get<OpenAIOption>();
// 文档切片配置
builder.Configuration.GetSection("TextChunker").Get<TextChunkerOption>();
// 配置数据库连接
builder.Configuration.GetSection("GraphDBConnection").Get<GraphDBConnectionOption>();
 
// 注意,需要先注入配置文件,然后再注入GraphRag.Net
builder.Services.AddGraphRagNet();

        这里,我们将在默认配置中注入OpenAI的配置、文本切片的配置和数据库连接的配置。然后,依次注入这些配置文件和GraphRag.Net的服务。

自定义配置方法

        如果需要自定义模型或本地模型,可能需要实现一些额外的服务接口,下面是自定义配置的示例:

1
2
3
4
5
6
7
var kernelBuild = Kernel.CreateBuilder();
kernelBuild.Services.AddKeyedSingleton<ITextGenerationService>("mock-text", new MockTextCompletion());
kernelBuild.Services.AddKeyedSingleton<IChatCompletionService>("mock-chat", new MockChatCompletion());
kernelBuild.Services.AddSingleton<ITextEmbeddingGenerationService>(new MockTextEmbeddingGeneratorService());
kernelBuild.Services.AddKeyedSingleton("mock-embedding", new MockTextEmbeddingGeneratorService());
 
builder.Services.AddGraphRagNet(kernelBuild.Build());

在这个自定义配置示例中,我们引入了三个自定义服务接口:ITextGenerationServiceIChatCompletionServiceITextEmbeddingGenerationService

实现自定义服务接口

接下来,我们需要为每个服务接口提供具体的实现。以下是三个接口的具体实现:

实现IChatCompletionService

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
public class MockChatCompletion : IChatCompletionService
{
    private readonly Dictionary<string, object?> _attributes = new();
    private string _chatId;
 
 
    private static readonly JsonSerializerOptions _jsonSerializerOptions = new()
    {
        NumberHandling = JsonNumberHandling.AllowReadingFromString,
        Encoder = JavaScriptEncoder.Create(UnicodeRanges.All)
    };
 
    public IReadOnlyDictionary<string, object?> Attributes => _attributes;
 
    public MockChatCompletion()
    {
 
    }
 
    public async Task<IReadOnlyList<ChatMessageContent>> GetChatMessageContentsAsync(ChatHistory chatHistory, PromptExecutionSettings? executionSettings = null, Kernel? kernel = null, [EnumeratorCancellation] CancellationToken cancellationToken = default)
    {
        StringBuilder sb = new();
        string result = $"这是一条Mock数据,便于聊天测试,你的消息是:{chatHistory.LastOrDefault().ToString()}";
        return [new(AuthorRole.Assistant, result.ToString())];
    }
 
    public async IAsyncEnumerable<StreamingChatMessageContent> GetStreamingChatMessageContentsAsync(ChatHistory chatHistory, PromptExecutionSettings? executionSettings = null, Kernel? kernel = null, [EnumeratorCancellation] CancellationToken cancellationToken = default)
    {
        StringBuilder sb = new();
        string result = $"这是一条Mock数据,便于聊天测试,你的消息是:{chatHistory.LastOrDefault().ToString()}";
        foreach (var c in result)
        {
            yield return new StreamingChatMessageContent(AuthorRole.Assistant, c.ToString());
        }
    }
}

  

实现ITextGenerationService

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
public class MockTextCompletion : ITextGenerationService, IAIService
{
    private readonly Dictionary<string, object?> _attributes = new();
    private string _chatId;
 
    private static readonly JsonSerializerOptions _jsonSerializerOptions = new()
    {
        NumberHandling = JsonNumberHandling.AllowReadingFromString,
        Encoder = JavaScriptEncoder.Create(UnicodeRanges.All)
    };
 
    public IReadOnlyDictionary<string, object?> Attributes => _attributes;
 
    public MockTextCompletion()
    {
 
    }
 
    public async Task<IReadOnlyList<TextContent>> GetTextContentsAsync(string prompt, PromptExecutionSettings? executionSettings = null, Kernel? kernel = null, CancellationToken cancellationToken = default)
    {
        StringBuilder sb = new();
        string result = $"这是一条Mock数据,便于聊天测试,你的消息是:{prompt}";
        return [new(result.ToString())];
    }
 
    public async IAsyncEnumerable<StreamingTextContent> GetStreamingTextContentsAsync(string prompt, PromptExecutionSettings? executionSettings = null, Kernel? kernel = null, CancellationToken cancellationToken = default)
    {
        StringBuilder sb = new();
        string result = $"这是一条Mock数据,便于聊天测试,你的消息是:{prompt}";
        foreach (var c in result)
        {
            var streamingTextContent = new StreamingTextContent(c.ToString(), modelId: "mock");
 
            yield return streamingTextContent;
        }
    }
}

实现ITextEmbeddingGenerationService

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
public sealed class MockTextEmbeddingGeneratorService : ITextEmbeddingGenerationService
{
    private Dictionary<string, object?> AttributesInternal { get; } = [];
    public IReadOnlyDictionary<string, object?> Attributes => this.AttributesInternal;
    public MockTextEmbeddingGeneratorService()
    {
 
    }
    public async Task<IList<ReadOnlyMemory<float>>> GenerateEmbeddingsAsync(
      IList<string> data,
      Kernel? kernel = null,
      CancellationToken cancellationToken = default)
    {
        IList<ReadOnlyMemory<float>> results = new List<ReadOnlyMemory<float>>();
 
        float[] array1 = { 1.0f, 2.0f, 3.0f };
        float[] array2 = { 4.0f, 5.0f, 6.0f };
        float[] array3 = { 7.0f, 8.0f, 9.0f };
 
        // 将数组包装为ReadOnlyMemory<float>并添加到列表中
        results.Add(new ReadOnlyMemory<float>(array1));
        results.Add(new ReadOnlyMemory<float>(array2));
        results.Add(new ReadOnlyMemory<float>(array3));
 
        return results;
    }
 
    public void Dispose()
    {
 
    }
}

  

        看到这里,你可能已经发现,集成自定义模型和本地模型非常简单。只需按照上述步骤,实现相应的接口并注入配置,你就可以在GraphRag.Net中使用这些自定义的功能。

 

结语

        通过本文的介绍,我们了解了如何在GraphRag.Net中集成国产模型和本地模型。希望大家能够根据这些示例,开发出更多适合自己需求的功能。更多精彩内容,欢迎关注我的公众号,并发送进群加入我们的GraphRag.Net交流群,与社区小伙伴们一起交流学习!

        感谢阅读,我们下期再见!

 

posted @   许泽宇  阅读(1011)  评论(0编辑  收藏  举报
相关博文:
阅读排行:
· 清华大学推出第四讲使用 DeepSeek + DeepResearch 让科研像聊天一样简单!
· 推荐几款开源且免费的 .NET MAUI 组件库
· 实操Deepseek接入个人知识库
· 易语言 —— 开山篇
· Trae初体验
点击右上角即可分享
微信分享提示