Qdrant向量資料庫練習及Azure AI Embedding API測試

float[] GetEmbedding(string text)
{
    //nuget: OpenAI-DotNet
    var endpoint = new Uri("https://{your name}.openai.azure.com/");
    var apiKey = "{your key}";
    var deploymentName = "text-embedding-3-small";

    var credential = new AzureKeyCredential(apiKey);
    AzureKeyCredential credentials = new(apiKey);

    AzureOpenAIClient azureOpenAIClient = new AzureOpenAIClient(endpoint, credentials);
    var embeddingClient = azureOpenAIClient.GetEmbeddingClient(deploymentName);

    var r = embeddingClient.GenerateEmbedding(text);
    var embedding = r.Value;
    ReadOnlyMemory<float> vec = embedding.ToFloats();
    return vec.ToArray();
}

async Task InitData()
{
    var client = new QdrantClient("localhost", 6334, false, "3065678qazwsx");
    string colName = "text_embedding";

    if (!await client.CollectionExistsAsync(colName))
    {
        await client.CreateCollectionAsync(colName,
            new VectorParams { Size = 1536, Distance = Distance.Cosine });
    }

    var count = await client.CountAsync(colName);


    List<TextData> data = new List<TextData>
    {
        new TextData{ catg = "布袋戲", text = "清香白蓮素還真" },
        new TextData{ catg = "布袋戲", text = "半神半聖亦半仙，全儒全道是全賢，腦中真書藏萬卷，掌握文武半邊天" },
        new TextData{ catg = "布袋戲", text = "霹靂化身最多的首席男主角，溫文儒雅、器宇軒昂、超凡脫俗、武學莫測高深、足智多謀、博學多能、謙虛有禮，處世圓融冷靜、慈悲親和、關懷眾生；以武林和平、天下大同為己任，『謀為天下謀、利為天下利』 - 無我、無為！為武林風塵默默承受一切，多次以絕頂智慧化解災厄，置之死地而後生，為天下蒼生應現各種精彩玄奇的身份！幽默風趣的隨機教化，難捨能捨、忍辱負重、不計毀謗、無怨無悔，默默付出，不遺餘力、不求回報，真乃具足大慈悲與大智慧的凡聖一體，反璞歸真『素還真』。" },
        new TextData{ catg = "布袋戲", text = "日月星三才子" },
        new TextData{ catg = "詩詞", text = "床前明月光，疑是地上霜。舉頭望明月，低頭思故鄉。"},
        new TextData{ catg = "詩詞", text = "春眠不覺曉，處處聞啼鳥。夜來風雨聲，花落知多少。"},
        new TextData{ catg = "詩詞", text = "滿室天香仙子家，一琴一劍一杯茶。羽衣常帶煙霞色，不惹人間桃李花。"},
    };

    var points = data.Select(t =>
    {
        var catg = t.catg;
        var text = t.text;
        PointId id = new PointId();
        id.Uuid = Guid.NewGuid().ToString();
        return new PointStruct
        {

            Id = id,
            Vectors = GetEmbedding(text),
            Payload =
            {
                    ["catg"] = catg,
                    ["text"] = text
            }
        };
    }).ToList();
    var updateResult = await client.UpsertAsync(colName, points);
}

var client = new QdrantClient("localhost", 6334, false, "3065678qazwsx");
string colName = "text_embedding";

//查詢語句轉換為向量
string keywd = "請隨機給我一首詩";
var queryVector = GetEmbedding(keywd);

Console.ForegroundColor = ConsoleColor.Cyan;
Console.WriteLine($"# 查詢：{keywd}");
Console.ForegroundColor = ConsoleColor.Yellow;
Console.WriteLine($"(向量相似度, TOP 3)");
Console.ResetColor();

var top3 = await client.SearchAsync(
  colName,
  queryVector,
  limit: 3);

int rank = 1;
foreach (var p in top3)
{
    Console.WriteLine($"{rank++}. {p.Score:n4} {p.Payload["catg"].StringValue} {p.Payload["text"].StringValue}");
}

Console.ForegroundColor = ConsoleColor.Yellow;
Console.WriteLine($"(向量相似度 + 限定詩布袋戲, TOP 3)");
Console.ResetColor();

queryVector = GetEmbedding("素還真是誰");
var top3poem = await client.SearchAsync(
  colName,
  queryVector,
  filter: Conditions.MatchText("catg", "布袋戲"),
  limit: 3);
rank = 1;
foreach (var p in top3poem)
{
    Console.WriteLine($"{rank++}. {p.Score:n4} {p.Payload["text"].StringValue}");
}