.Net: Increase auto-invoke and in-flight tool calling hard-coded limi…

…ts (#6272)

As we discussed offline yesterday, with auto function calling filters,
someone can now put their own limits in place, so raising these
hard-coded back stops to something much less likely to be hit. We could
subsequently get rid of them completely if desired.

---------

Co-authored-by: SergeyMenshykh <68852919+SergeyMenshykh@users.noreply.github.com>
Co-authored-by: SergeyMenshykh <sergemenshikh@gmail.com>

dotnet/src/Connectors/Connectors.Google.UnitTests/GeminiToolCallBehaviorTests.cs

@@ -30,11 +30,12 @@ public void EnableKernelFunctionsReturnsCorrectKernelFunctionsInstance()
     public void AutoInvokeKernelFunctionsReturnsCorrectKernelFunctionsInstance()
     {
         // Arrange & Act
+        const int DefaultMaximumAutoInvokeAttempts = 128;
         var behavior = GeminiToolCallBehavior.AutoInvokeKernelFunctions;
 
         // Assert
         Assert.IsType<GeminiToolCallBehavior.KernelFunctions>(behavior);
-        Assert.Equal(5, behavior.MaximumAutoInvokeAttempts);
+        Assert.Equal(DefaultMaximumAutoInvokeAttempts, behavior.MaximumAutoInvokeAttempts);
     }
 
     [Fact]

dotnet/src/Connectors/Connectors.Google/Core/Gemini/Clients/GeminiChatCompletionClient.cs

@@ -46,7 +46,7 @@ internal sealed class GeminiChatCompletionClient : ClientBase
     /// was invoked with), but we do want to limit it. This limit is arbitrary and can be tweaked in the future and/or made
     /// configurable should need arise.
     /// </remarks>
-    private const int MaxInflightAutoInvokes = 5;
+    private const int MaxInflightAutoInvokes = 128;
 
     /// <summary>Tracking <see cref="AsyncLocal{Int32}"/> for <see cref="MaxInflightAutoInvokes"/>.</summary>
     private static readonly AsyncLocal<int> s_inflightAutoInvokes = new();

dotnet/src/Connectors/Connectors.Google/GeminiToolCallBehavior.cs

@@ -32,7 +32,7 @@ public abstract class GeminiToolCallBehavior
     /// support, where the model can request multiple tools in a single response, it is significantly
     /// less likely that this limit is reached, as most of the time only a single request is needed.
     /// </remarks>
-    private const int DefaultMaximumAutoInvokeAttempts = 5;
+    private const int DefaultMaximumAutoInvokeAttempts = 128;
 
     /// <summary>
     /// Gets an instance that will provide all of the <see cref="Kernel"/>'s plugins' function information.

dotnet/src/Connectors/Connectors.OpenAI/AzureSdk/ClientCore.cs

@@ -49,7 +49,7 @@ internal abstract class ClientCore
     /// was invoked with), but we do want to limit it. This limit is arbitrary and can be tweaked in the future and/or made
     /// configurable should need arise.
     /// </remarks>
-    private const int MaxInflightAutoInvokes = 5;
+    private const int MaxInflightAutoInvokes = 128;
 
     /// <summary>Singleton tool used when tool call count drops to 0 but we need to supply tools to keep the service happy.</summary>
     private static readonly ChatCompletionsFunctionToolDefinition s_nonInvocableFunctionTool = new() { Name = "NonInvocableTool" };

dotnet/src/Connectors/Connectors.OpenAI/ToolCallBehavior.cs

@@ -36,7 +36,7 @@ public abstract class ToolCallBehavior
     /// support, where the model can request multiple tools in a single response, it is significantly
     /// less likely that this limit is reached, as most of the time only a single request is needed.
     /// </remarks>
-    private const int DefaultMaximumAutoInvokeAttempts = 5;
+    private const int DefaultMaximumAutoInvokeAttempts = 128;
 
     /// <summary>
     /// Gets an instance that will provide all of the <see cref="Kernel"/>'s plugins' function information.

dotnet/src/Connectors/Connectors.UnitTests/OpenAI/ChatCompletion/AzureOpenAIChatCompletionServiceTests.cs

@@ -323,8 +323,8 @@ public async Task GetChatMessageContentsWithFunctionCallAsync()
     public async Task GetChatMessageContentsWithFunctionCallMaximumAutoInvokeAttemptsAsync()
     {
         // Arrange
-        const int DefaultMaximumAutoInvokeAttempts = 5;
-        const int AutoInvokeResponsesCount = 6;
+        const int DefaultMaximumAutoInvokeAttempts = 128;
+        const int ModelResponsesCount = 129;
 
         int functionCallCount = 0;
 
@@ -342,7 +342,7 @@ public async Task GetChatMessageContentsWithFunctionCallMaximumAutoInvokeAttempt
 
         var responses = new List<HttpResponseMessage>();
 
-        for (var i = 0; i < AutoInvokeResponsesCount; i++)
+        for (var i = 0; i < ModelResponsesCount; i++)
         {
             responses.Add(new HttpResponseMessage(HttpStatusCode.OK) { Content = new StringContent(OpenAITestHelper.GetTestResponse("chat_completion_single_function_call_test_response.json")) });
         }
@@ -501,8 +501,8 @@ public async Task GetStreamingChatMessageContentsWithFunctionCallAsync()
     public async Task GetStreamingChatMessageContentsWithFunctionCallMaximumAutoInvokeAttemptsAsync()
     {
         // Arrange
-        const int DefaultMaximumAutoInvokeAttempts = 5;
-        const int AutoInvokeResponsesCount = 6;
+        const int DefaultMaximumAutoInvokeAttempts = 128;
+        const int ModelResponsesCount = 129;
 
         int functionCallCount = 0;
 
@@ -520,7 +520,7 @@ public async Task GetStreamingChatMessageContentsWithFunctionCallMaximumAutoInvo
 
         var responses = new List<HttpResponseMessage>();
 
-        for (var i = 0; i < AutoInvokeResponsesCount; i++)
+        for (var i = 0; i < ModelResponsesCount; i++)
         {
             responses.Add(new HttpResponseMessage(HttpStatusCode.OK) { Content = new StringContent(OpenAITestHelper.GetTestResponse("chat_completion_streaming_single_function_call_test_response.txt")) });
         }

dotnet/src/Connectors/Connectors.UnitTests/OpenAI/ToolCallBehaviorTests.cs

@@ -30,11 +30,12 @@ public void EnableKernelFunctionsReturnsCorrectKernelFunctionsInstance()
     public void AutoInvokeKernelFunctionsReturnsCorrectKernelFunctionsInstance()
     {
         // Arrange & Act
+        const int DefaultMaximumAutoInvokeAttempts = 128;
         var behavior = ToolCallBehavior.AutoInvokeKernelFunctions;
 
         // Assert
         Assert.IsType<KernelFunctions>(behavior);
-        Assert.Equal(5, behavior.MaximumAutoInvokeAttempts);
+        Assert.Equal(DefaultMaximumAutoInvokeAttempts, behavior.MaximumAutoInvokeAttempts);
     }
 
     [Fact]