ReadmeTests.cs

using LangChain.Databases;
using LangChain.Databases.InMemory;
using LangChain.Docstore;
using LangChain.Indexes;
using LangChain.Providers.OpenAI;
using LangChain.Sources;
using LangChain.TextSplitters;
using LangChain.VectorStores;
using static LangChain.Chains.Chain;

namespace LangChain.IntegrationTests;

[TestFixture]
public class ReadmeTests
{
    [Explicit]
    [Test]
    public async Task Readme()
    {
        var gpt4 = new Gpt4Model(
            Environment.GetEnvironmentVariable("OPENAI_API_KEY") ??
            throw new InconclusiveException("OPENAI_API_KEY is not set"));
        var index = await InMemoryVectorStore.CreateIndexFromDocuments(gpt4, new[]
        {
            "I spent entire day watching TV",
            "My dog name is Bob",
            "This ice cream is delicious",
            "It is cold in space"
        }.ToDocuments());

        var chain = (
            Set("What is the good name for a pet?", outputKey: "question") |
            RetrieveDocuments(index, inputKey: "question", outputKey: "documents") |
            StuffDocuments(inputKey: "documents", outputKey: "context") |
            Template("""
                Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer.

                {context}

                Question: {question}
                Helpful Answer:
                """, outputKey: "prompt") |
            LLM(gpt4, inputKey: "prompt", outputKey: "pet_sentence")) >>
            Template("""
                Human will provide you with sentence about pet. You need to answer with pet name.

                Human: My dog name is Jack
                Answer: Jack
                Human: I think the best name for a pet is "Jerry"
                Answer: Jerry
                Human: {pet_sentence}
                Answer:
                """, outputKey: "prompt") |
            LLM(gpt4, inputKey: "prompt", outputKey: "text");

        var result = await chain.Run(resultKey: "text");
        result.Should().Be("Bob");
    }
    
    /// <summary>
    /// Price to run from zero(create embedding and request to LLM): 0,015$
    /// Price to re-run if database is exists: 0,0004$
    /// </summary>
    /// <exception cref="InconclusiveException"></exception>
    [Explicit]
    [Test]
    public async Task RagWithOpenAiUsingChains()
    {
        var gpt35 = new Gpt35TurboModel(
            Environment.GetEnvironmentVariable("OPENAI_API_KEY") ??
            throw new InconclusiveException("OPENAI_API_KEY is not set"));
        
        // Create database with embeddings from Harry Potter book pdf
        if (!File.Exists("vectors.db"))
        {
            await using var stream = H.Resources.HarryPotterBook1_pdf.AsStream();
            var documents = await PdfPigPdfSource.FromStreamAsync(stream);
            
            await SQLiteVectorStore.CreateIndexFromDocuments(
                embeddings: gpt35,
                documents: documents,
                filename: "vectors.db",
                tableName: "vectors",
                textSplitter: new RecursiveCharacterTextSplitter(
                    chunkSize: 200,
                    chunkOverlap: 50));
        }

        var vectorStore = new SQLiteVectorStore("vectors.db", "vectors", gpt35);
        var index = new VectorStoreIndexWrapper(vectorStore);
        var chain =
            // set the question
            Set("Who was drinking a unicorn blood?", outputKey: "question") |
            // take 5 most similar documents
            RetrieveSimilarDocuments(index, inputKey: "question", outputKey: "documents", amount: 5) |
            // combine documents together and put them into context
            CombineDocuments(inputKey: "documents", outputKey: "context") |
            // replace context and question in the prompt with their values
            Template(@"Use the following pieces of context to answer the question at the end.
If the answer is not in context then just say that you don't know, don't try to make up an answer.
Keep the answer as short as possible.

{context}

Question: {question}
Helpful Answer:") |
            // send the result to the language model
            LargeLanguageModel(gpt35);

        var result = await chain.Run("text");
        
        Console.WriteLine($"LLM answer: {result}");
        Console.WriteLine($"Total usage: {gpt35.TotalUsage}");
    }
    
    /// <summary>
    /// Price to run from zero(create embeddings and request to LLM): 0,015$
    /// Price to re-run if database is exists: 0,0004$
    /// </summary>
    /// <exception cref="InconclusiveException"></exception>
    [Explicit]
    [Test]
    public async Task RagWithOpenAiUsingAsyncMethods()
    {
        var gpt35 = new Gpt35TurboModel(
            Environment.GetEnvironmentVariable("OPENAI_API_KEY") ??
            throw new InconclusiveException("OPENAI_API_KEY is not set"));
        
        if (!File.Exists("vectors.db"))
        {
            var documents = await PdfPigPdfSource.FromUriAsync(
                new Uri("https://canonburyprimaryschool.co.uk/wp-content/uploads/2016/01/Joanne-K.-Rowling-Harry-Potter-Book-1-Harry-Potter-and-the-Philosophers-Stone-EnglishOnlineClub.com_.pdf"));
            
            await SQLiteVectorStore.CreateIndexFromDocuments(
                embeddings: gpt35,
                documents: documents,
                filename: "vectors.db",
                tableName: "vectors",
                textSplitter: new RecursiveCharacterTextSplitter(
                    chunkSize: 200,
                    chunkOverlap: 50));
        }

        var database = new SQLiteVectorStore(
            filename: "vectors.db",
            tableName: "vectors",
            embeddings: gpt35);
        const string question = "Who was drinking a unicorn blood?";
        var similarDocuments = await database.GetSimilarDocuments(question, amount: 5);
        
        var answer = await gpt35.GenerateAsync(
            $"""
             Use the following pieces of context to answer the question at the end.
             If the answer is not in context then just say that you don't know, don't try to make up an answer.
             Keep the answer as short as possible.

             {similarDocuments.AsString()}

             Question: {question}
             Helpful Answer:
             """, CancellationToken.None).ConfigureAwait(false);
        
        Console.WriteLine($"LLM answer: {answer}"); // The cloaked figure.
        Console.WriteLine($"Total usage: {gpt35.TotalUsage}");
    }

    [Explicit]
    [Test]
    public async Task SimpleDocuments()
    {
        // https://www.together.ai/blog/embeddings-endpoint-release
        // var together = new OpenAiModel(new OpenAiConfiguration
        // {
        //     ApiKey = Environment.GetEnvironmentVariable("TOGETHER_API_KEY") ??
        //              throw new InconclusiveException("TOGETHER_API_KEY is not set"),
        //     Endpoint = "api.together.xyz",
        //     ModelId = "togethercomputer/Qwen-7B",
        //     EmbeddingModelId = "togethercomputer/m2-bert-80M-32k-retrieval",
        // });
        var gpt35 = new Gpt35TurboModel(
            Environment.GetEnvironmentVariable("OPENAI_API_KEY") ??
            throw new InconclusiveException("OPENAI_API_KEY is not set"));
        
        var database = await InMemoryVectorStore.CreateIndexFromDocuments(gpt35, new[]
        {
            "I spent entire day watching TV",
            "My dog name is Bob",
            "This ice cream is delicious",
            "It is cold in space"
        }.ToDocuments());

        const string question = "What is the good name for a pet?";
        var similarDocuments = await database.Store.GetSimilarDocuments(question, amount: 1);
        
        var petNameResponse = await gpt35.GenerateAsync(
            $"""

             Human will provide you with sentence about pet. You need to answer with pet name.

             Human: My dog name is Jack
             Answer: Jack
             Human: I think the best name for a pet is "Jerry"
             Answer: Jerry
             Human: {similarDocuments.AsString()}
             Answer:
             """, CancellationToken.None).ConfigureAwait(false);
        
        Console.WriteLine($"LLM answer: {petNameResponse}");
        Console.WriteLine($"Total usage: {gpt35.TotalUsage}");
        
        petNameResponse.ToString().Should().Be("Bob");
    }
}