Azure Open IA, acelerando nossos caminhos.

Agora provides the programs you need to speed up your infrastructure using the .NET Core console. The program performs tasks related to Azure Open AI, composes messages, and verifies basic checks by respondents against primary documents. Para isso, utilizaremos o pacote Azure.AI.OpenAI. There are no specific examples. There are cases where version 2.0.0 is used.

install-package Azure.AI.OpenAI

Primeiro passo vamos criar um client da classe AzureOpenAIClient:

AzureOpenAIClient azureClient = new(endpoint, credential);

This is an eo endpoint with credentials, which can be used repeatedly across instances of Azure OpenAI by referring to the “Keys and endpoints” section.

 var credential = new AzureKeyCredential("...");

Depois, vamos criar uma instancia da classe chamada Search Client. Essa classe ajudará na conexão com o serviço de busca (search service).

You know and know:

var searchEndpoint = new Uri("...");
var searchCredential = new AzureKeyCredential("...");
var indexName = "gptkbindex";
var searchClient = new SearchClient(searchEndpoint, indexName, searchCredential);
var searchOptions = new SearchOptions
{
    Size = 5 // Número de documentos a recuperar
};
var searchResults = searchClient.Search(text, searchOptions);
var retrievedDocuments = searchResults.Value.GetResults().Select(result => result.Document["content"].ToString());
var context = string.Join("\n", retrievedDocuments);

A class Search Client What you need is uma chave de segurança. Para encontrá-la, acessamos a instância do serviço de busca (Search Service) followed by “Keys”.

o Endpoint Assessment Overview

Agora vamos criar uma instancia da classe ChatClientI requested a response for OpenAI and passed the prompt for Azure openIA.

E aqui que está o Grande Segredo do RAGtudo é prompt, pegamos o resultado da busca do Search Client I think there are various variables context e juntamos has no prompts.

var prompt = $"Contexto: {context}\nPergunta: {text}";
ChatClient chatClient = await azureClient.CompleteChatAsync("gpt-4o");
ChatCompletion completion = chatClient.CompleteChat(new List
{
new SystemChatMessage("Você é um assistente atencioso"),
new UserChatMessage(prompt),

});

Para obter o valor do campo deployment namePlease don’t describe gpt-4o or OpenAI instances. Summarize “Model Deployment” and see “Managed Deployment”. Isso abrirá o Azure OpenAI StudioThere are also ways to visualize the implanted model.

os pacotes utilizes the forums:

using Azure;
using Azure.AI.OpenAI;
using Azure.Search.Documents;
using Azure.Search.Documents.Models;
using OpenAI.Chat;


public class Program
{

    static async Task Main(string[] args)
    {
        while (true)
        {
            Console.WriteLine("Digite uma Pergunta");
            var question = Console.ReadLine();
            if (question != null)
            {
                var result = await AskingChatCompletionWithSearchsAsync(question);
                Console.WriteLine(result);
            }

        }

    }

    static async Task AskingChatCompletionWithSearchsAsync(string text)
    {

        var endpoint = new Uri("https://openiapriv02.openai.azure.com/");
        var credential = new AzureKeyCredential("...");
        AzureOpenAIClient azureClient = new(endpoint, credential);


        var searchEndpoint = new Uri("https://gptkb-kv4atymcdg6pg.search.windows.net");
        var searchCredential = new AzureKeyCredential("...");
        var indexName = "gptkbindex";
        var searchClient = new SearchClient(searchEndpoint, indexName, searchCredential);
        var searchOptions = new SearchOptions
        {
            Size = 5 // Número de documentos a recuperar
        };
        var searchResults = searchClient.Search(text, searchOptions);
        var retrievedDocuments = searchResults.Value.GetResults().Select(result => result.Document["content"].ToString());
        var context = string.Join("\n", retrievedDocuments);



        var prompt = $"Contexto: {context}\nPergunta: {text}";
        ChatClient chatClient = azureClient.GetChatClient("gpt-4o");
        ChatCompletion completion = await chatClient.CompleteChatAsync(new List
        {
            new SystemChatMessage("Você é um assistente atencioso"),
            new UserChatMessage(prompt),

        });


        var result = $"{completion.Role}: {completion.Content[0].Text}";
        return result;
    }
}

fiz uma pergunta para o support com base em um documento que já veio de exemplo no acelerador o Benefit_Options.pdf, perguntei o Is there a Northwind Health Plus program? Observe que a resposta edada com base nos documentsos indexados no search service

This model continues to work with new characters, and there are also versions of the model that can be controlled by OpenAI.

Aqui temos o uso de um conector e de um plugin, o connector vai adicionar accesso ao IA Search por meio dessa Implementação:

var memoryBuilder = new MemoryBuilder();
memoryBuilder.WithMemoryStore(new AzureAISearchMemoryStore(
   "https://gptkb-nggxeq6fpjnxg.search.windows.net",
   "...")
);

Alem do Pacote do Microsoft.SemanticKernel Ainda vamos precisar de mais dois pacotes o Microsoft.SemanticKernel.Connectors.AzureAISearch uh Microsoft.SemanticKernel.Plugins.Memory

Mas também precisamos do plugin Text memory plugin para acessar as memorias, mas ates de acessar vamos cria-las

Para isso vamos usar Essa Implementação:

var memory = memoryBuilder.Build();  
  
const string MemoryCollectionName = "aboutMe";  
  
await memory.SaveInformationAsync(MemoryCollectionName, id: "info1", text: "My name is Andrea");  
await memory.SaveInformationAsync(MemoryCollectionName, id: "info2", text: "I currently work as a tourist operator");  
await memory.SaveInformationAsync(MemoryCollectionName, id: "info3", text: "I currently live in Seattle and have been living there since 2005");  
await memory.SaveInformationAsync(MemoryCollectionName, id: "info4", text: "I visited France and Italy five times since 2015");  
await memory.SaveInformationAsync(MemoryCollectionName, id: "info5", text: "My family is from New York");

kernel.ImportPluginFromObject(new TextMemoryPlugin(memory));

Com isso sera gerado um index na IA search chamado aboutme

list index

search explorer

Campos do index

esse index foi gerado usando esse Text Embedding text-embedding-ada-002que é um tipo de algoritmo de vetorização

Deploy Azure OpenAI Studio

com explains how to implement an instance configuration for Azure openIA or include ferntas ferntas for the text-embedding-ada-002 model.

 memoryBuilder.WithTextEmbeddingGeneration((loggerFactory, httpClient) => {
     return new AzureOpenAITextEmbeddingGenerationService(
         "text-embedding-ada-002", // Embedding generation service name
         "https://openiapriv02.openai.azure.com/",
         "...",
         httpClient: httpClient,
         loggerFactory: loggerFactory
     );
 });

por fim para fazer as perguntas ao chat e receber resposta que vem desse index como se fossem uma lembrança do chat, mas antes precisamos enriquecer o prompt com essas informações algo assim:

 const string skPrompt = @"  
 ChatBot can have a conversation with you about any topic.  
 It can give explicit instructions or say 'I don't know' if it does not have an answer.  
  
 Information about me, from previous conversations:  
 - {{$fact1}} {{recall $fact1}}  
 - {{$fact2}} {{recall $fact2}}  
 - {{$fact3}} {{recall $fact3}}  
 - {{$fact4}} {{recall $fact4}}  
 - {{$fact5}} {{recall $fact5}}  
  
 Chat:  
 {{$history}}  
 User: {{$userInput}}  
 ChatBot: ";

esses valores serão passados ​​​​pelo KernelArguments:

 arguments\["fact1"\] = "What my name?";  
 arguments\["fact2"\] = "where do I live?";  
 arguments\["fact3"\] = "where is my family from?";  
 arguments\["fact4"\] = "where have I travelled?";  
 arguments\["fact5"\] = "what do I do for work?";

Check out the complete implementation:

var builder = Kernel.CreateBuilder();

builder.AddAzureOpenAIChatCompletion(
         "gpt-35-turbo",                      // Azure OpenAI Deployment Name
         "https://cog-nggxeq6fpjnxg.openai.azure.com/", // Azure OpenAI Endpoint
         "...");  // Azure OpenAI Key


var kernel = builder.Build();

#pragma warning disable SKEXP0001, SKEXP0010, SKEXP0050, SKEXP0020

var memoryBuilder = new MemoryBuilder();


memoryBuilder.WithTextEmbeddingGeneration((loggerFactory, httpClient) => {
    return new AzureOpenAITextEmbeddingGenerationService(
        "text-embedding-ada-002", // Embedding generation service name
        "https://openiapriv02.openai.azure.com/",
        "...",
        httpClient: httpClient,
        loggerFactory: loggerFactory
    );
});


memoryBuilder.WithMemoryStore(new AzureAISearchMemoryStore(
   "https://gptkb-nggxeq6fpjnxg.search.windows.net",
   "...")
);

var memory = memoryBuilder.Build();

const string MemoryCollectionName = "aboutMe";

await memory.SaveInformationAsync(MemoryCollectionName, id: "info1", text: "My name is Andrea");
await memory.SaveInformationAsync(MemoryCollectionName, id: "info2", text: "I currently work as a tourist operator");
await memory.SaveInformationAsync(MemoryCollectionName, id: "info3", text: "I currently live in Seattle and have been living there since 2005");
await memory.SaveInformationAsync(MemoryCollectionName, id: "info4", text: "I visited France and Italy five times since 2015");
await memory.SaveInformationAsync(MemoryCollectionName, id: "info5", text: "My family is from New York");

var questions = new[]
{
    "what is my name?",
    "where do I live?",
    "where is my family from?",
    "where have I travelled?",
    "what do I do for work?",
};


#pragma warning disable SKEXP0050

// TextMemoryPlugin provides the "recall" function
kernel.ImportPluginFromObject(new TextMemoryPlugin(memory));


const string skPrompt = @"
ChatBot can have a conversation with you about any topic.
It can give explicit instructions or say 'I don't know' if it does not have an answer.

Information about me, from previous conversations:
- {{$fact1}} {{recall $fact1}}
- {{$fact2}} {{recall $fact2}}
- {{$fact3}} {{recall $fact3}}
- {{$fact4}} {{recall $fact4}}
- {{$fact5}} {{recall $fact5}}

Chat:
{{$history}}
User: {{$userInput}}
ChatBot: ";

var chatFunction = kernel.CreateFunctionFromPrompt(skPrompt, new OpenAIPromptExecutionSettings { MaxTokens = 200, Temperature = 0.8 });

#pragma warning disable SKEXP0050

var arguments = new KernelArguments();

arguments["fact1"] = "What my name?";
arguments["fact2"] = "where do I live?";
arguments["fact3"] = "where is my family from?";
arguments["fact4"] = "where have I travelled?";
arguments["fact5"] = "what do I do for work?";

arguments[TextMemoryPlugin.CollectionParam] = MemoryCollectionName;
arguments[TextMemoryPlugin.LimitParam] = "2";
arguments[TextMemoryPlugin.RelevanceParam] = "0.8";

var history = "";
arguments["history"] = history;
Func Chat = async (string input) => {
    // Save new message in the kernel arguments
    arguments["userInput"] = input;

    // Process the user message and get an answer
    var answer = await chatFunction.InvokeAsync(kernel, arguments);

    // Append the new interaction to the chat history
    var result = $"\nUser: {input}\nChatBot: {answer}\n";

    history += result;
    arguments["history"] = history;

    // Show the bot response
    Console.WriteLine(result);
};


while (true)
{
    Console.WriteLine("Digite uma Pergunta");
    var question = Console.ReadLine();
    if (question != null)
    {
        await Chat(question);
    }

}

The console and execution response is a response to “What my name”?

Usage: