feat: Refactor providers

LangChain.sln

@@ -172,11 +172,13 @@ Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "LangChain.Providers.Automat
 EndProject
 Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "LangChain.Providers.Automatic1111.IntegrationTests", "src\tests\LangChain.Providers.Automatic1111.IntegrationTests\LangChain.Providers.Automatic1111.IntegrationTests.csproj", "{A6CF79BC-8365-46E8-9230-1A4AD615D40B}"
 EndProject
+Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "LangChain.Providers.Abstractions", "src\libs\Providers\LangChain.Providers.Abstractions\LangChain.Providers.Abstractions.csproj", "{628DDC9D-28A1-4C2F-BA13-171BD8CF711C}"
+EndProject
 Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "LangChain.Samples.FileMemory", "examples\LangChain.Samples.FileMemory\LangChain.Samples.FileMemory.csproj", "{BA701280-0BEB-4DA4-92B3-9C777082C2AF}"
 EndProject
-Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "LangChain.Providers.Bedrock.IntegrationTests", "src\tests\LangChain.Providers.Bedrock.IntegrationTests\LangChain.Providers.Bedrock.IntegrationTests.csproj", "{73C76E80-95C5-4C96-A319-4F32043C903E}"
+Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "LangChain.Providers.Amazon.Bedrock.IntegrationTests", "src\tests\LangChain.Providers.Amazon.Bedrock.IntegrationTests\LangChain.Providers.Amazon.Bedrock.IntegrationTests.csproj", "{73C76E80-95C5-4C96-A319-4F32043C903E}"
 EndProject
-Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "LangChain.Providers.Bedrock", "src\libs\Providers\LangChain.Providers.Bedrock\LangChain.Providers.Bedrock.csproj", "{67985CCB-F606-41F8-9D36-513459F58882}"
+Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "LangChain.Providers.Amazon.Bedrock", "src\libs\Providers\LangChain.Providers.Amazon.Bedrock\LangChain.Providers.Amazon.Bedrock.csproj", "{67985CCB-F606-41F8-9D36-513459F58882}"
 EndProject
 Global
 	GlobalSection(SolutionConfigurationPlatforms) = preSolution
@@ -408,6 +410,10 @@ Global
 		{A6CF79BC-8365-46E8-9230-1A4AD615D40B}.Debug|Any CPU.Build.0 = Debug|Any CPU
 		{A6CF79BC-8365-46E8-9230-1A4AD615D40B}.Release|Any CPU.ActiveCfg = Release|Any CPU
 		{A6CF79BC-8365-46E8-9230-1A4AD615D40B}.Release|Any CPU.Build.0 = Release|Any CPU
+		{628DDC9D-28A1-4C2F-BA13-171BD8CF711C}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
+		{628DDC9D-28A1-4C2F-BA13-171BD8CF711C}.Debug|Any CPU.Build.0 = Debug|Any CPU
+		{628DDC9D-28A1-4C2F-BA13-171BD8CF711C}.Release|Any CPU.ActiveCfg = Release|Any CPU
+		{628DDC9D-28A1-4C2F-BA13-171BD8CF711C}.Release|Any CPU.Build.0 = Release|Any CPU
 		{BA701280-0BEB-4DA4-92B3-9C777082C2AF}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
 		{BA701280-0BEB-4DA4-92B3-9C777082C2AF}.Debug|Any CPU.Build.0 = Debug|Any CPU
 		{BA701280-0BEB-4DA4-92B3-9C777082C2AF}.Release|Any CPU.ActiveCfg = Release|Any CPU
@@ -488,6 +494,7 @@ Global
 		{4913844F-74EC-4E74-AE8A-EA825569E6BA} = {E55391DE-F8F3-4CC2-A0E3-2406C76E9C68}
 		{BF4C7B87-0997-4208-84EF-D368DF7B9861} = {E55391DE-F8F3-4CC2-A0E3-2406C76E9C68}
 		{A6CF79BC-8365-46E8-9230-1A4AD615D40B} = {FDEE2E22-C239-4921-83B2-9797F765FD6A}
+		{628DDC9D-28A1-4C2F-BA13-171BD8CF711C} = {E55391DE-F8F3-4CC2-A0E3-2406C76E9C68}
 		{BA701280-0BEB-4DA4-92B3-9C777082C2AF} = {F17A86AE-A174-4B6C-BAA7-9D9A9704BE85}
 		{73C76E80-95C5-4C96-A319-4F32043C903E} = {FDEE2E22-C239-4921-83B2-9797F765FD6A}
 		{67985CCB-F606-41F8-9D36-513459F58882} = {E55391DE-F8F3-4CC2-A0E3-2406C76E9C68}

examples/LangChain.Samples.HuggingFace/Program.cs

@@ -1,5 +1,4 @@
-using LangChain.Providers;
-using LangChain.Providers.HuggingFace;
+using LangChain.Providers.HuggingFace.Predefined;
 
 using var client = new HttpClient();
 var gpt2Model = new Gpt2Model(apiKey: string.Empty, client);

examples/LangChain.Samples.OpenAI/Program.cs

@@ -1,5 +1,4 @@
-using LangChain.Providers;
-using LangChain.Providers.OpenAI;
+using LangChain.Providers.OpenAI.Predefined;
 
 var apiKey =
     Environment.GetEnvironmentVariable("OPENAI_API_KEY") ??

src/Directory.Packages.props

@@ -47,6 +47,7 @@
     <PackageVersion Include="StackExchange.Redis" Version="2.7.4" />
     <PackageVersion Include="System.Net.Http" Version="4.3.4" />
     <PackageVersion Include="System.Text.Json" Version="8.0.0" />
+    <PackageVersion Include="System.ValueTuple" Version="4.5.0" />
     <PackageVersion Include="Tiktoken" Version="1.1.3" />
     <PackageVersion Include="tryAGI.OpenAI" Version="2.0.0-alpha.9" />
     <PackageVersion Include="Microsoft.Extensions.DependencyInjection.Abstractions" Version="8.0.0" />

src/libs/Databases/LangChain.Databases.Chroma/ChromaVectorStore.cs

@@ -1,7 +1,7 @@
 using System.Text.Json;
-using LangChain.Abstractions.Embeddings.Base;
 using LangChain.Common.Converters;
 using LangChain.Docstore;
+using LangChain.Providers;
 using LangChain.VectorStores;
 using Microsoft.SemanticKernel.AI.Embeddings;
 using Microsoft.SemanticKernel.Connectors.Memory.Chroma;
@@ -31,7 +31,7 @@ public class ChromaVectorStore : VectorStore
     public ChromaVectorStore(
         HttpClient httpClient,
         string endpoint,
-        IEmbeddings embeddings,
+        IEmbeddingModel embeddings,
         string collectionName = LangchainDefaultCollectionName)
         : base(embeddings)
     {
@@ -166,8 +166,8 @@ public override async Task<IEnumerable<Document>> SimilaritySearchByVectorAsync(
         int k = 4,
         CancellationToken cancellationToken = default)
     {
-        var embeddings = await Embeddings
-            .EmbedQueryAsync(query, cancellationToken)
+        var embeddings = await EmbeddingModel
+            .CreateEmbeddingsAsync(query, null, cancellationToken)
             .ConfigureAwait(false);
 
         var documentsWithScores = await SimilaritySearchByVectorWithAsync(embeddings, k, cancellationToken).ConfigureAwait(false);
@@ -192,8 +192,8 @@ public override async Task<IEnumerable<Document>> MaxMarginalRelevanceSearch(
         float lambdaMult = 0.5f,
         CancellationToken cancellationToken = default)
     {
-        var embeddings = await Embeddings
-            .EmbedQueryAsync(query, cancellationToken)
+        float[] embeddings = await EmbeddingModel
+            .CreateEmbeddingsAsync(query, null, cancellationToken)
             .ConfigureAwait(false);
 
         var documents = await MaxMarginalRelevanceSearchByVector(
@@ -251,8 +251,8 @@ private async Task<IEnumerable<string>> AddCoreAsync(
         string[] ids,
         CancellationToken cancellationToken)
     {
-        var embeddings = await Embeddings
-            .EmbedDocumentsAsync(texts, cancellationToken)
+        float[][] embeddings = await EmbeddingModel
+            .CreateEmbeddingsAsync(texts, null, cancellationToken)
             .ConfigureAwait(false);
 
         var records = new MemoryRecord[texts.Length];

src/libs/Databases/LangChain.Databases.InMemory/InMemoryVectorStore.cs

@@ -1,6 +1,6 @@
-using LangChain.Abstractions.Embeddings.Base;
-using LangChain.Docstore;
+using LangChain.Docstore;
 using LangChain.Indexes;
+using LangChain.Providers;
 using LangChain.TextSplitters;
 using LangChain.VectorStores;
 
@@ -15,7 +15,7 @@ namespace LangChain.Databases.InMemory;
 /// <param name="embeddings"></param>
 /// <param name="distanceMetrics"></param>
 public class InMemoryVectorStore(
-    IEmbeddings embeddings,
+    IEmbeddingModel embeddings,
     EDistanceMetrics distanceMetrics = EDistanceMetrics.Euclidean)
     : VectorStore(embeddings)
 {
@@ -33,7 +33,7 @@ public class InMemoryVectorStore(
     /// <param name="documents"></param>
     /// <returns></returns>
     public static async Task<VectorStoreIndexWrapper> CreateIndexFromDocuments(
-        IEmbeddings embeddings,
+        IEmbeddingModel embeddings,
         IReadOnlyCollection<Document> documents)
     {
         var vectorStore = new InMemoryVectorStore(embeddings);
@@ -51,9 +51,9 @@ public override async Task<IEnumerable<string>> AddDocumentsAsync(
     {
         var docs = documents.ToArray();
 
-        var embeddings = await Embeddings.EmbedDocumentsAsync(docs
+        float[][] embeddings = await EmbeddingModel.CreateEmbeddingsAsync(docs
             .Select(x => x.PageContent)
-            .ToArray(), cancellationToken).ConfigureAwait(false);
+            .ToArray(), cancellationToken: cancellationToken).ConfigureAwait(false);
         var ids = new List<string>();
         for (var i = 0; i < docs.Length; i++)
         {
@@ -98,9 +98,9 @@ public override async Task<IEnumerable<Document>> SimilaritySearchAsync(
         int k = 4,
         CancellationToken cancellationToken = default)
     {
-        var embedding = await Embeddings.EmbedQueryAsync(
+        float[] embedding = await EmbeddingModel.CreateEmbeddingsAsync(
             query,
-            cancellationToken).ConfigureAwait(false);
+            cancellationToken: cancellationToken).ConfigureAwait(false);
 
         return await SimilaritySearchByVectorAsync(
             embedding,
@@ -129,15 +129,14 @@ public override Task<IEnumerable<Document>> SimilaritySearchByVectorAsync(
         int k = 4,
         CancellationToken cancellationToken = default)
     {
-        var embedding = await Embeddings.EmbedQueryAsync(
+        float[] embedding = await EmbeddingModel.CreateEmbeddingsAsync(
             query,
-            cancellationToken).ConfigureAwait(false);
-        var arr = embedding.ToArray();
+            cancellationToken: cancellationToken).ConfigureAwait(false);
         var distances = _storage.Select(s =>
             new
             {
                 doc = s.doc,
-                distance = _distanceFunction(s.vec, arr)
+                distance = _distanceFunction(s.vec, embedding)
             }).Take(k);
 
         return distances.Select(d => new ValueTuple<Document, float>(d.doc, d.distance));

src/libs/Databases/LangChain.Databases.Postgres/PostgresVectorStore.cs

@@ -1,6 +1,6 @@
-using LangChain.Abstractions.Embeddings.Base;
 using LangChain.Databases.Postgres;
 using LangChain.Docstore;
+using LangChain.Providers;
 using LangChain.VectorStores;
 
 namespace LangChain.Databases;
@@ -25,12 +25,12 @@ public class PostgresVectorStore : VectorStore
     public PostgresVectorStore(
         string connectionString,
         int vectorSize,
-        IEmbeddings embeddings,
+        IEmbeddingModel embeddingModel,
         string schema = DefaultSchema,
         string collectionName = DefaultCollectionName,
         DistanceStrategy distanceStrategy = DistanceStrategy.Cosine,
         Func<float, float>? overrideRelevanceScoreFn = null)
-        : base(embeddings, overrideRelevanceScoreFn)
+        : base(embeddingModel, overrideRelevanceScoreFn)
     {
         _distanceStrategy = distanceStrategy;
         _collectionName = collectionName;
@@ -44,8 +44,10 @@ public override async Task<IEnumerable<string>> AddDocumentsAsync(
         CancellationToken cancellationToken = default)
     {
         var documentsArray = documents.ToArray();
-        var embeddings = await Embeddings
-            .EmbedDocumentsAsync(documentsArray.Select(d => d.PageContent).ToArray(), cancellationToken)
+        float[][] embeddings = await EmbeddingModel
+            .CreateEmbeddingsAsync(documentsArray
+                .Select(d => d.PageContent)
+                .ToArray(), null, cancellationToken)
             .ConfigureAwait(false);
 
         var ids = new string[documentsArray.Length];
@@ -75,8 +77,8 @@ public override async Task<IEnumerable<string>> AddTextsAsync(
         var textsArray = texts.ToArray();
         var metadatasArray = metadatas?.ToArray() ?? new Dictionary<string, object>?[textsArray.Length];
 
-        var embeddings = await Embeddings
-            .EmbedDocumentsAsync(textsArray, cancellationToken)
+        float[][] embeddings = await EmbeddingModel
+            .CreateEmbeddingsAsync(textsArray, null, cancellationToken)
             .ConfigureAwait(false);
 
         var ids = new string[textsArray.Length];
@@ -137,9 +139,11 @@ public override async Task<IEnumerable<Document>> SimilaritySearchByVectorAsync(
         string query, int k = 4,
         CancellationToken cancellationToken = default)
     {
-        var embedding = await Embeddings.EmbedQueryAsync(query, cancellationToken).ConfigureAwait(false);
+        float[] embedding = await EmbeddingModel.CreateEmbeddingsAsync(
+            query, null, cancellationToken).ConfigureAwait(false);
 
-        return await SimilaritySearchByVectorWithScoreAsync(embedding, k, cancellationToken)
+        return await SimilaritySearchByVectorWithScoreAsync(
+                embedding, k, cancellationToken)
             .ConfigureAwait(false);
     }
 
@@ -183,7 +187,7 @@ public override async Task<IEnumerable<Document>> MaxMarginalRelevanceSearch(
         int fetchK = 20, float lambdaMult = 0.5f,
         CancellationToken cancellationToken = default)
     {
-        var embedding = await Embeddings.EmbedQueryAsync(query, cancellationToken).ConfigureAwait(false);
+        float[] embedding = await EmbeddingModel.CreateEmbeddingsAsync(query, null, cancellationToken).ConfigureAwait(false);
 
         return await MaxMarginalRelevanceSearchByVector(embedding, k, fetchK, lambdaMult, cancellationToken)
             .ConfigureAwait(false);

src/libs/Databases/LangChain.Databases.Sqlite/SQLiteVectorStore.cs

@@ -1,8 +1,8 @@
 using System.Text.Json;
-using LangChain.Abstractions.Embeddings.Base;
 using LangChain.Base;
 using LangChain.Docstore;
 using LangChain.Indexes;
+using LangChain.Providers;
 using LangChain.TextSplitters;
 using LangChain.VectorStores;
 using Microsoft.Data.Sqlite;
@@ -28,7 +28,7 @@ public sealed class SQLiteVectorStore : VectorStore, IDisposable
     /// <param name="textSplitter"></param>
     /// <returns></returns>
     public static async Task<VectorStoreIndexWrapper> CreateIndexFromDocuments(
-        IEmbeddings embeddings,
+        IEmbeddingModel embeddings,
         IReadOnlyCollection<Document> documents,
         string filename = "vectorstore.db",
         string tableName = "vectors",
@@ -41,18 +41,20 @@ public static async Task<VectorStoreIndexWrapper> CreateIndexFromDocuments(
         return index;
     }
 
-    public static SQLIteVectorStoreOptions DefaultOptions = new SQLIteVectorStoreOptions();
+    public static SQLIteVectorStoreOptions DefaultOptions { get; } = new();
 
     /// <summary>
     ///  If database does not exists, it loads documents from the documentsSource, creates an index from these documents and returns the created index.
     ///  If database exists, it loads the index from the database.
     ///  documentsSource is used only if the database does not exist. If the database exists, documentsSource is ignored.
     /// </summary>
-    /// <param name="embeddings">An object implementing the <see cref="IEmbeddings"/> interface. This object is used to generate embeddings for the documents.</param>
+    /// <param name="embeddings">An object implementing the <see cref="IEmbeddingModel"/> interface. This object is used to generate embeddings for the documents.</param>
     /// <param name="documentsSource">An optional object implementing the <see cref="ISource"/> interface. This object is used to load documents if the vector store database file does not exist.</param>
     /// <param name="options">An optional <see cref="SQLIteVectorStoreOptions"/> object. This object provides configuration options for the SQLite vector store</param>
     public static async Task<VectorStoreIndexWrapper> GetIndex(
-        IEmbeddings embeddings, ISource? documentsSource=null, SQLIteVectorStoreOptions? options=null)
+        IEmbeddingModel embeddings,
+        ISource? documentsSource = null,
+        SQLIteVectorStoreOptions? options = null)
     {
         options ??= DefaultOptions;
 
@@ -61,10 +63,11 @@ public static async Task<VectorStoreIndexWrapper> GetIndex(
 
         if (!System.IO.File.Exists("vectors.db"))
         {
-            var documents = await documentsSource.LoadAsync();
-            return await SQLiteVectorStore.CreateIndexFromDocuments(embeddings, documents, options.Filename, options.TableName, textSplitter: textSplitter);
-        }
+            var documents = await documentsSource.LoadAsync().ConfigureAwait(false);
 
+            return await CreateIndexFromDocuments(
+                embeddings, documents, options.Filename, options.TableName, textSplitter: textSplitter).ConfigureAwait(false);
+        }
 
         var vectorStore = new SQLiteVectorStore(options.Filename, options.TableName, embeddings);
         var index = new VectorStoreIndexWrapper(vectorStore);
@@ -81,7 +84,7 @@ public static async Task<VectorStoreIndexWrapper> GetIndex(
     public SQLiteVectorStore(
         string filename,
         string tableName,
-        IEmbeddings embeddings,
+        IEmbeddingModel embeddings,
         EDistanceMetrics distanceMetrics = EDistanceMetrics.Euclidean)
         : base(embeddings)
     {
@@ -201,7 +204,11 @@ public override async Task<IEnumerable<string>> AddDocumentsAsync(
 
         var docs = documents.ToArray();
 
-        var embeddings = await Embeddings.EmbedDocumentsAsync(docs.Select(x => x.PageContent).ToArray(), cancellationToken).ConfigureAwait(false);
+        float[][] embeddings = await EmbeddingModel.CreateEmbeddingsAsync(
+            docs
+                .Select(x => x.PageContent)
+                .ToArray(),
+            cancellationToken: cancellationToken).ConfigureAwait(false);
         List<string> ids = new List<string>();
         for (int i = 0; i < docs.Length; i++)
         {
@@ -258,10 +265,19 @@ public override async Task<bool> DeleteAsync(IEnumerable<string> ids, Cancellati
     /// <param name="k"></param>
     /// <param name="cancellationToken"></param>
     /// <returns></returns>
-    public override async Task<IEnumerable<Document>> SimilaritySearchAsync(string query, int k = 4, CancellationToken cancellationToken = default)
+    public override async Task<IEnumerable<Document>> SimilaritySearchAsync(
+        string query,
+        int k = 4,
+        CancellationToken cancellationToken = default)
     {
-        var embedding = await Embeddings.EmbedQueryAsync(query, cancellationToken).ConfigureAwait(false);
-        return await SimilaritySearchByVectorAsync(embedding, k, cancellationToken).ConfigureAwait(false);
+        float[] embedding = await EmbeddingModel.CreateEmbeddingsAsync(
+            query,
+            cancellationToken: cancellationToken).ConfigureAwait(false);
+
+        return await SimilaritySearchByVectorAsync(
+            embedding,
+            k,
+            cancellationToken).ConfigureAwait(false);
     }
 
     /// <summary>
@@ -289,9 +305,13 @@ public override async Task<IEnumerable<Document>> SimilaritySearchByVectorAsync(
     public override async Task<IEnumerable<(Document, float)>> SimilaritySearchWithScoreAsync(string query,
         int k = 4, CancellationToken cancellationToken = default)
     {
-        var embedding = await Embeddings.EmbedQueryAsync(query, cancellationToken).ConfigureAwait(false);
-        var arr = embedding.ToArray();
-        var documents = await SearchByVector(arr, k).ConfigureAwait(false);
+        float[] embedding = await EmbeddingModel.CreateEmbeddingsAsync(
+            query,
+            cancellationToken: cancellationToken).ConfigureAwait(false);
+        var documents = await SearchByVector(
+            embedding,
+            k).ConfigureAwait(false);
+
         return documents;
     }