.Net: Add GenAI support to Connectors.Onnx (microsoft#6518)

@stephentoub invite me to Contribute
https://github.com/feiyun0112/SemanticKernel.Connectors.OnnxRuntimeGenAI
to Microsoft.SemanticKernel.Connectors.Onnx


[https://github.com/feiyun0112/SemanticKernel.Connectors.OnnxRuntimeGenAI/issues/4](https://github.com/feiyun0112/SemanticKernel.Connectors.OnnxRuntimeGenAI/issues/4)

---------

Co-authored-by: Mark Wallace <127216156+markwallace-microsoft@users.noreply.github.com>
Co-authored-by: Stephen Toub <stoub@microsoft.com>
Co-authored-by: Roger Barreto <19890735+RogerBarreto@users.noreply.github.com>

dotnet/Directory.Packages.props

@@ -130,5 +130,9 @@
       <PrivateAssets>all</PrivateAssets>
       <IncludeAssets>runtime; build; native; contentfiles; analyzers; buildtransitive</IncludeAssets>
     </PackageReference>
+    <!-- OnnxRuntimeGenAI -->
+    <PackageVersion Include="Microsoft.ML.OnnxRuntimeGenAI" Version="0.3.0"/>
+    <PackageVersion Include="Microsoft.ML.OnnxRuntimeGenAI.Cuda" Version="0.3.0"/>
+	  <PackageVersion Include="Microsoft.ML.OnnxRuntimeGenAI.DirectML" Version="0.3.0"/>
   </ItemGroup>
 </Project>

dotnet/src/Connectors/Connectors.Onnx.UnitTests/OnnxExtensionsTests.cs

@@ -0,0 +1,49 @@
+// Copyright (c) Microsoft. All rights reserved.
+
+using Microsoft.Extensions.DependencyInjection;
+using Microsoft.SemanticKernel;
+using Microsoft.SemanticKernel.ChatCompletion;
+using Microsoft.SemanticKernel.Connectors.Onnx;
+using Xunit;
+
+namespace SemanticKernel.Connectors.Onnx.UnitTests;
+
+/// <summary>
+/// Unit tests for <see cref="OnnxKernelBuilderExtensions"/>.
+/// </summary>
+public class OnnxExtensionsTests
+{
+    [Fact]
+    public void AddOnnxRuntimeGenAIChatCompletionToServiceCollection()
+    {
+        // Arrange
+        var collection = new ServiceCollection();
+        collection.AddOnnxRuntimeGenAIChatCompletion("modelId", "modelPath");
+
+        // Act
+        var kernelBuilder = collection.AddKernel();
+        var kernel = collection.BuildServiceProvider().GetRequiredService<Kernel>();
+        var service = kernel.GetRequiredService<IChatCompletionService>();
+
+        // Assert
+        Assert.NotNull(service);
+        Assert.IsType<OnnxRuntimeGenAIChatCompletionService>(service);
+    }
+
+    [Fact]
+    public void AddOnnxRuntimeGenAIChatCompletionToKernelBuilder()
+    {
+        // Arrange
+        var collection = new ServiceCollection();
+        var kernelBuilder = collection.AddKernel();
+        kernelBuilder.AddOnnxRuntimeGenAIChatCompletion("modelId", "modelPath");
+
+        // Act
+        var kernel = collection.BuildServiceProvider().GetRequiredService<Kernel>();
+        var service = kernel.GetRequiredService<IChatCompletionService>();
+
+        // Assert
+        Assert.NotNull(service);
+        Assert.IsType<OnnxRuntimeGenAIChatCompletionService>(service);
+    }
+}

dotnet/src/Connectors/Connectors.Onnx.UnitTests/OnnxRuntimeGenAIPromptExecutionSettingsTests.cs

@@ -0,0 +1,97 @@
+// Copyright (c) Microsoft. All rights reserved.
+
+using System.Text.Json;
+using Microsoft.SemanticKernel;
+using Microsoft.SemanticKernel.Connectors.Onnx;
+using Xunit;
+
+namespace SemanticKernel.Connectors.Onnx.UnitTests;
+
+/// <summary>
+/// Unit tests for <see cref="OnnxRuntimeGenAIPromptExecutionSettings"/>.
+/// </summary>
+public class OnnxRuntimeGenAIPromptExecutionSettingsTests
+{
+    [Fact]
+    public void FromExecutionSettingsWhenAlreadyMistralShouldReturnSame()
+    {
+        // Arrange
+        var executionSettings = new OnnxRuntimeGenAIPromptExecutionSettings();
+
+        // Act
+        var onnxExecutionSettings = OnnxRuntimeGenAIPromptExecutionSettings.FromExecutionSettings(executionSettings);
+
+        // Assert
+        Assert.Same(executionSettings, onnxExecutionSettings);
+    }
+
+    [Fact]
+    public void FromExecutionSettingsWhenNullShouldReturnDefaultSettings()
+    {
+        // Arrange
+        PromptExecutionSettings? executionSettings = null;
+
+        // Act
+        var onnxExecutionSettings = OnnxRuntimeGenAIPromptExecutionSettings.FromExecutionSettings(executionSettings);
+
+        // Assert        
+        Assert.Null(onnxExecutionSettings.TopK);
+        Assert.Null(onnxExecutionSettings.TopP);
+        Assert.Null(onnxExecutionSettings.Temperature);
+        Assert.Null(onnxExecutionSettings.RepetitionPenalty);
+        Assert.Null(onnxExecutionSettings.PastPresentShareBuffer);
+        Assert.Null(onnxExecutionSettings.NumReturnSequences);
+        Assert.Null(onnxExecutionSettings.NumBeams);
+        Assert.Null(onnxExecutionSettings.NoRepeatNgramSize);
+        Assert.Null(onnxExecutionSettings.MinTokens);
+        Assert.Null(onnxExecutionSettings.MaxTokens);
+        Assert.Null(onnxExecutionSettings.LengthPenalty);
+        Assert.Null(onnxExecutionSettings.DiversityPenalty);
+        Assert.Null(onnxExecutionSettings.EarlyStopping);
+        Assert.Null(onnxExecutionSettings.DoSample);
+    }
+
+    [Fact]
+    public void FromExecutionSettingsWhenSerializedHasPropertiesShouldPopulateSpecialized()
+    {
+        // Arrange
+        string jsonSettings = """
+                                {
+                                    "top_k": 2,
+                                    "top_p": 0.9,
+                                    "temperature": 0.5,
+                                    "repetition_penalty": 0.1,
+                                    "past_present_share_buffer": true,
+                                    "num_return_sequences": 200,
+                                    "num_beams": 20,
+                                    "no_repeat_ngram_size": 15,
+                                    "min_tokens": 10,
+                                    "max_tokens": 100,
+                                    "length_penalty": 0.2,
+                                    "diversity_penalty": 0.3,
+                                    "early_stopping": false,
+                                    "do_sample": true
+                                }
+                                """;
+
+        // Act
+        var executionSettings = JsonSerializer.Deserialize<PromptExecutionSettings>(jsonSettings);
+        var onnxExecutionSettings = OnnxRuntimeGenAIPromptExecutionSettings.FromExecutionSettings(executionSettings);
+
+        // Assert
+        Assert.Equal(2, onnxExecutionSettings.TopK);
+        Assert.Equal(0.9f, onnxExecutionSettings.TopP);
+        Assert.Equal(0.5f, onnxExecutionSettings.Temperature);
+        Assert.Equal(0.1f, onnxExecutionSettings.RepetitionPenalty);
+        Assert.True(onnxExecutionSettings.PastPresentShareBuffer);
+        Assert.Equal(200, onnxExecutionSettings.NumReturnSequences);
+        Assert.Equal(20, onnxExecutionSettings.NumBeams);
+        Assert.Equal(15, onnxExecutionSettings.NoRepeatNgramSize);
+        Assert.Equal(10, onnxExecutionSettings.MinTokens);
+        Assert.Equal(100, onnxExecutionSettings.MaxTokens);
+        Assert.Equal(0.2f, onnxExecutionSettings.LengthPenalty);
+        Assert.Equal(0.3f, onnxExecutionSettings.DiversityPenalty);
+        Assert.False(onnxExecutionSettings.EarlyStopping);
+        Assert.True(onnxExecutionSettings.DoSample);
+    }
+}

dotnet/src/Connectors/Connectors.Onnx/Connectors.Onnx.csproj

@@ -25,4 +25,10 @@
     <PackageReference Include="System.Numerics.Tensors" />
   </ItemGroup>
 
+  <ItemGroup>
+    <PackageReference Include="Microsoft.ML.OnnxRuntimeGenAI" Condition=" '$(Configuration)' == 'Debug' OR '$(Configuration)' == 'Release' " />
+    <PackageReference Include="Microsoft.ML.OnnxRuntimeGenAI.Cuda" Condition=" '$(Configuration)' == 'Debug_Cuda' OR '$(Configuration)' == 'Release_Cuda' " />
+    <PackageReference Include="Microsoft.ML.OnnxRuntimeGenAI.DirectML" Condition=" '$(Configuration)' == 'Debug_DirectML' OR '$(Configuration)' == 'Release_DirectML' " />    
+  </ItemGroup>
+
 </Project>

dotnet/src/Connectors/Connectors.Onnx/OnnxKernelBuilderExtensions.cs

@@ -2,6 +2,8 @@
 
 using System.IO;
 using Microsoft.Extensions.DependencyInjection;
+using Microsoft.Extensions.Logging;
+using Microsoft.SemanticKernel.ChatCompletion;
 using Microsoft.SemanticKernel.Connectors.Onnx;
 using Microsoft.SemanticKernel.Embeddings;
 
@@ -14,6 +16,29 @@ namespace Microsoft.SemanticKernel;
 /// </summary>
 public static class OnnxKernelBuilderExtensions
 {
+    /// <summary>
+    /// Add OnnxRuntimeGenAI Chat Completion services to the kernel builder.
+    /// </summary>
+    /// <param name="builder">The kernel builder.</param>
+    /// <param name="modelId">Model Id.</param>
+    /// <param name="modelPath">The generative AI ONNX model path.</param>
+    /// <param name="serviceId">The optional service ID.</param>
+    /// <returns>The updated kernel builder.</returns>
+    public static IKernelBuilder AddOnnxRuntimeGenAIChatCompletion(
+        this IKernelBuilder builder,
+        string modelId,
+        string modelPath,
+        string? serviceId = null)
+    {
+        builder.Services.AddKeyedSingleton<IChatCompletionService>(serviceId, (serviceProvider, _) =>
+            new OnnxRuntimeGenAIChatCompletionService(
+                modelId,
+                modelPath: modelPath,
+                loggerFactory: serviceProvider.GetService<ILoggerFactory>()));
+
+        return builder;
+    }
+
     /// <summary>Adds a text embedding generation service using a BERT ONNX model.</summary>
     /// <param name="builder">The <see cref="IKernelBuilder"/> instance to augment.</param>
     /// <param name="onnxModelPath">The path to the ONNX model file.</param>