[C#] Expose cached Generator through IChatClient.GetService

src/csharp/OnnxRuntimeGenAIChatClient.cs

@@ -246,6 +246,11 @@ generator.ConversationId is null ||
     }
 
     /// <inheritdoc/>
+    /// <remarks>
+    /// A <see cref="Generator"/> service is available only while a generator is cached between requests.
+    /// The returned generator is owned by this client and must not be disposed, retained across requests,
+    /// or used concurrently with a request.
+    /// </remarks>
     object? IChatClient.GetService(Type serviceType, object? serviceKey)
     {
         if (serviceType is null)
@@ -259,6 +264,7 @@ generator.ConversationId is null ||
             serviceType == typeof(Model) ? _model :
             serviceType == typeof(Tokenizer) ? _tokenizer :
             serviceType == typeof(Config) ? _config :
+            serviceType == typeof(Generator) ? Volatile.Read(ref _cachedGenerator)?.Generator :
             serviceType?.IsInstanceOfType(this) is true ? this :
             null;
     }
@@ -418,4 +424,4 @@ private sealed class YieldAwaiter : INotifyCompletion
         public void UnsafeOnCompleted(Action continuation) => Task.Run(continuation);
         public void GetResult() { }
     }
-}
+}

test/csharp/TestOnnxRuntimeGenAIAPI.cs

@@ -538,6 +538,31 @@ public async Task TestChatClient()
             Assert.NotEmpty(completion.Text);
         }
 
+        [Fact(DisplayName = "TestChatClientGetServiceGenerator")]
+        public async Task TestChatClientGetServiceGenerator()
+        {
+            OnnxRuntimeGenAIChatClientOptions options = new()
+            {
+                EnableCaching = true,
+                PromptFormatter = static (messages, _) => string.Concat(messages.Select(message => message.Text)),
+            };
+
+            using var client = new OnnxRuntimeGenAIChatClient(_tinyRandomGpt2ModelPath, options);
+            IChatClient chatClient = client;
+
+            Assert.Null(chatClient.GetService<Generator>(null));
+
+            await client.GetResponseAsync("H", new()
+            {
+                MaxOutputTokens = 1,
+                Temperature = 0f,
+            });
+
+            var generator = Assert.IsType<Generator>(chatClient.GetService<Generator>(null));
+            Assert.Same(generator, chatClient.GetService<Generator>(null));
+            Assert.Null(chatClient.GetService<Generator>(new object()));
+        }
+
         [IgnoreOnModelAbsenceFact(DisplayName = "TestTokenizerBatchEncodeDecode")]
         public void TestTokenizerBatchEncodeDecode()
         {