feat: Sequential chain with QnA (#52)

* VectorStoreIndexCreator, error fixes, tweeking, Documents QnA test

* changing Web and Pdf sources to use propper document

* bugfixes and massive simplification of DocumentQnA test

* bugfix

* Sequential chain with QnA

src/libs/LangChain.Core/Indexes/VectorStoreIndexWrapper.cs

@@ -9,19 +9,19 @@ namespace LangChain.Indexes;
 
 public class VectorStoreIndexWrapper
 {
-    private readonly VectorStore _vectorStore;
+    public VectorStore Store { get; }
 
     public VectorStoreIndexWrapper(VectorStore vectorStore)
     {
-        _vectorStore = vectorStore;
+        Store = vectorStore;
     }
 
     public Task<string?> QueryAsync(string question, BaseCombineDocumentsChain llm, string inputKey= "question", string outputKey= "output_text")
     {
         var chain = new RetrievalQaChain(
             new RetrievalQaChainInput(
                 llm,
-                _vectorStore.AsRetreiver())
+                Store.AsRetreiver())
             {
                 InputKey= inputKey,
                 OutputKey= outputKey,

src/tests/LangChain.Providers.LLamaSharp.IntegrationTests/LLamaSharpTests.Helpers.cs

@@ -0,0 +1,60 @@
+using LangChain.Abstractions.Embeddings.Base;
+using LangChain.Databases.InMemory;
+using LangChain.Docstore;
+using LangChain.Indexes;
+using LangChain.Prompts;
+using LangChain.TextSplitters;
+
+namespace LangChain.Providers.LLamaSharp.IntegrationTests;
+
+public partial class LLamaSharpTests
+{
+
+    IEmbeddings CreateEmbeddings()
+    {
+        var embeddings = new LLamaSharpEmbeddings(new LLamaSharpConfiguration
+        {
+            PathToModelFile = ModelPath,
+            Temperature = 0
+        });
+        return embeddings;
+
+    }
+
+    IChatModel CreateInstructionModel()
+    {
+        var model = new LLamaSharpModelInstruction(new LLamaSharpConfiguration
+        {
+            PathToModelFile = ModelPath,
+            Temperature = 0
+        });
+        return model;
+
+    }
+    IChatModel CreateChatModel()
+    {
+        var model = new LLamaSharpModelChat(new LLamaSharpConfiguration
+        {
+            PathToModelFile = ModelPath,
+            Temperature = 0
+        });
+        return model;
+
+    }
+    VectorStoreIndexWrapper CreateVectorStoreIndex(IEmbeddings embeddings, string[] texts)
+    {
+        InMemoryVectorStore vectorStore = new InMemoryVectorStore(embeddings);
+        var textSplitter = new CharacterTextSplitter();
+        VectorStoreIndexCreator indexCreator = new VectorStoreIndexCreator(vectorStore, textSplitter);
+        var index = indexCreator.FromDocumentsAsync(texts.Select(x => new Document(x)).ToList()).Result;
+        return index;
+    }
+
+    PromptTemplate CreatePromptTemplate()
+    {
+        string prompt = "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.\n\n{context}\n\nQuestion: {question}\nHelpful Answer:";
+        var template = new PromptTemplate(new PromptTemplateInput(prompt, new List<string>() { "context", "question" }));
+        return template;
+    }
+
+}

src/tests/LangChain.Providers.LLamaSharp.IntegrationTests/LLamaSharpTests.cs

@@ -1,6 +1,9 @@
-using LangChain.Abstractions.Embeddings.Base;
+using LangChain.Abstractions.Chains.Base;
+using LangChain.Abstractions.Embeddings.Base;
 using LangChain.Chains.CombineDocuments;
 using LangChain.Chains.LLM;
+using LangChain.Chains.RetrievalQA;
+using LangChain.Chains.Sequentials;
 using LangChain.Databases;
 using LangChain.Databases.InMemory;
 using LangChain.Docstore;
@@ -15,7 +18,7 @@
 namespace LangChain.Providers.LLamaSharp.IntegrationTests;
 
 [TestClass]
-public class LLamaSharpTests
+public partial class LLamaSharpTests
 {
     string ModelPath=>HuggingFaceModelDownloader.Instance.GetModel("TheBloke/Thespis-13B-v0.5-GGUF", "thespis-13b-v0.5.Q2_K.gguf","main").Result;
     [TestMethod]
@@ -106,48 +109,6 @@ public void EmbeddingsTestWithInMemory()
         Assert.AreEqual("My dog name is Bob", closest.PageContent);
     }
 
-
-    #region Helpers
-    IEmbeddings CreateEmbeddings()
-    {
-        var embeddings = new LLamaSharpEmbeddings(new LLamaSharpConfiguration
-        {
-            PathToModelFile = ModelPath,
-            Temperature = 0
-        });
-        return embeddings;
-
-    }
-
-    IChatModel CreateInstructionModel()
-    {
-        var model = new LLamaSharpModelInstruction(new LLamaSharpConfiguration
-        {
-            PathToModelFile = ModelPath,
-            Temperature = 0
-        });
-        return model;
-
-    }
-
-    VectorStoreIndexWrapper CreateVectorStoreIndex(IEmbeddings embeddings, string[] texts)
-    {
-        InMemoryVectorStore vectorStore = new InMemoryVectorStore(embeddings);
-        var textSplitter = new CharacterTextSplitter();
-        VectorStoreIndexCreator indexCreator = new VectorStoreIndexCreator(vectorStore, textSplitter);
-        var index = indexCreator.FromDocumentsAsync(texts.Select(x => new Document(x)).ToList()).Result;
-        return index;
-    }
-
-    PromptTemplate CreatePromptTemplate()
-    {
-        string prompt = "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.\n\n{context}\n\nQuestion: {question}\nHelpful Answer:";
-        var template = new PromptTemplate(new PromptTemplateInput(prompt, new List<string>() { "context", "question" }));
-        return template;
-    }
-    #endregion
-
-
     [TestMethod]
 #if CONTINUOUS_INTEGRATION_BUILD
     [Ignore]
@@ -186,10 +147,77 @@ public void DocumentsQuestionAnsweringTest()
             ).Result;
 
 
-
-
         Assert.IsTrue(answer.Contains("Bob"));
     }
 
+    IChain CreateChain1(IChatModel model,IEmbeddings embeddings)
+    {
+
+        string[] texts = new string[]
+        {
+            "I spent entire day watching TV",
+            "My dog name is Bob",
+            "This icecream is delicious",
+            "It is cold in space"
+        };
+
+        var index = CreateVectorStoreIndex(embeddings, texts);
+        var template = CreatePromptTemplate();
+
+        var llmchain = new LlmChain(new LlmChainInput(model, template)
+        {
+            OutputKey = "pet_sentence",
+        });
 
+        var stuffDocumentsChain = new StuffDocumentsChain(new StuffDocumentsChainInput(llmchain)
+        {
+            DocumentVariableName = "context",
+
+        });
+
+        var chain = new RetrievalQaChain(
+            new RetrievalQaChainInput(
+                stuffDocumentsChain,
+                index.Store.AsRetreiver())
+            {
+                InputKey = "question",
+                OutputKey = "pet_sentence",
+            }
+        );
+
+        return chain;
+    }
+
+    [TestMethod]
+#if CONTINUOUS_INTEGRATION_BUILD
+    [Ignore]
+#endif
+    public void SequentialChainTest()
+    {
+        // setup
+        var embeddings = CreateEmbeddings();
+        var model = CreateInstructionModel();
+
+        var chain1 = CreateChain1(model, embeddings);
+
+        var prompt =
+            @"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:";
+        var template = new PromptTemplate(new PromptTemplateInput(prompt, new List<string>() { "pet_sentence" }));
+        var chain2 = new LlmChain(new LlmChainInput(model, template));
+
+        var sequence = new SequentialChain(
+            new SequentialChainInput(
+                new [] { chain1, chain2 },
+                inputVariables:new[]{"question"}));
+
+        var answer=sequence.Run("What is the good name for a pet?").Result;
+
+        Assert.AreEqual("Bob", answer);
+    }
 }