Update Azure.AI.OpenAI to 2.1.0 and add tests

CHANGELOG.md

@@ -24,6 +24,10 @@ Starting v0.1.0 for Microsoft.Azure.WebJobs.Extensions.OpenAI.AzureAISearch, it
 - Updated Azure.AI.OpenAI from 1.0.0-beta.15 to 2.1.0
 - Updated Azure.Data.Tables from 12.9.1 to 12.10.0, Azure.Identity from 1.12.1 to 1.13.2, Microsoft.Extensions.Azure from 1.7.5 to 1.10.0
 
+### Added
+
+- Introduced experimental `isReasoningModel` property to support reasoning models. Setting of max_completion_tokens and reasoning_effort is not supported with current underlying Azure.AI.OpenAI
+
 ## v0.18.0 - 2024/10/08
 
 ### Changed

java-library/CHANGELOG.md

@@ -19,6 +19,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - Update azure-ai-openai from 1.0.0-beta.11 to 1.0.0-beta.16
 
+### Added
+
+- Introduced experimental `isReasoningModel` property to support reasoning models. Setting of max_completion_tokens and reasoning_effort is not supported with current underlying Azure.AI.OpenAI
+
 ## v0.4.0 - 2024/10/08
 
 ### Changed

java-library/src/main/java/com/microsoft/azure/functions/openai/annotation/assistant/AssistantPost.java

@@ -130,4 +130,11 @@
      * @return The maxTokens value.
      */
     String maxTokens() default "100";
+
+    /**
+     * Indicates whether the assistant uses a reasoning model.
+     *
+     * @return {@code true} if the assistant is based on a reasoning model; {@code false} otherwise.
+     */
+    boolean isReasoningModel() default false;
 }

java-library/src/main/java/com/microsoft/azure/functions/openai/annotation/search/SemanticSearch.java

@@ -138,4 +138,11 @@ String systemPrompt() default "You are a helpful assistant. You are responding t
      * @return The maxTokens value.
      */
     String maxTokens() default "2048";
+
+    /**
+     * Indicates whether the assistant uses a reasoning model.
+     *
+     * @return {@code true} if the assistant is based on a reasoning model; {@code false} otherwise.
+     */
+    boolean isReasoningModel();
 }

java-library/src/main/java/com/microsoft/azure/functions/openai/annotation/textcompletion/TextCompletion.java

@@ -91,4 +91,10 @@
      */
     String maxTokens() default "100";
 
+    /**
+     * Indicates whether the assistant uses a reasoning model.
+     *
+     * @return {@code true} if the assistant is based on a reasoning model; {@code false} otherwise.
+     */
+    boolean isReasoningModel();
 }

src/Functions.Worker.Extensions.OpenAI/Assistants/AssistantPostInputAttribute.cs

@@ -93,4 +93,12 @@ public AssistantPostInputAttribute(string id, string userMessage)
     /// Most models have a context length of 2048 tokens (except for the newest models, which support 4096).
     /// </remarks>
     public string? MaxTokens { get; set; } = "100";
+
+    /// <summary>
+    /// Gets or sets a value indicating whether the model is a reasoning model.
+    /// </summary>
+    /// <remarks>
+    /// Warning: This is experimental and associated with the reasoning model until all models have parity in the expected properties.
+    /// </remarks>
+    public bool IsReasoningModel { get; set; }
 }

src/Functions.Worker.Extensions.OpenAI/Search/SemanticSearchInputAttribute.cs

@@ -143,4 +143,12 @@ The following is a list of documents that you can refer to when answering questi
     /// Most models have a context length of 2048 tokens (except for the newest models, which support 4096).
     /// </remarks>
     public string? MaxTokens { get; set; } = "2048";
+
+    /// <summary>
+    /// Gets or sets a value indicating whether the model is a reasoning model.
+    /// </summary>
+    /// <remarks>
+    /// Warning: This is experimental and associated with the reasoning model until all models have parity in the expected properties.
+    /// </remarks>
+    public bool IsReasoningModel { get; set; }
 }

src/Functions.Worker.Extensions.OpenAI/TextCompletion/TextCompletionInputAttribute.cs

@@ -73,4 +73,12 @@ public TextCompletionInputAttribute(string prompt)
     /// Most models have a context length of 2048 tokens (except for the newest models, which support 4096).
     /// </remarks>
     public string? MaxTokens { get; set; } = "100";
+
+    /// <summary>
+    /// Gets or sets a value indicating whether the model is a reasoning model.
+    /// </summary>
+    /// <remarks>
+    /// Warning: This is experimental and associated with the reasoning model until all models have parity in the expected properties.
+    /// </remarks>
+    public bool IsReasoningModel { get; set; }
 }

src/WebJobs.Extensions.OpenAI/Assistants/AssistantBaseAttribute.cs

@@ -63,6 +63,14 @@ public class AssistantBaseAttribute : Attribute
     [AutoResolve]
     public string? TopP { get; set; }
 
+    /// <summary>
+    /// Gets or sets a value indicating whether the model is a reasoning model.
+    /// </summary>
+    /// <remarks>
+    /// Warning: This is experimental and associated with the reasoning model until all models have parity in the expected properties.
+    /// </remarks>
+    public bool IsReasoningModel { get; set; }
+
     /// <summary>
     /// Gets or sets the maximum number of tokens to output in the completion. Default value = 100.
     /// </summary>
@@ -76,21 +84,39 @@ public class AssistantBaseAttribute : Attribute
     internal ChatCompletionOptions BuildRequest()
     {
         ChatCompletionOptions request = new();
-        if (int.TryParse(this.MaxTokens, out int maxTokens))
+        if (float.TryParse(this.TopP, out float topP))
         {
-            request.MaxOutputTokenCount = maxTokens;
+            request.TopP = topP;
         }
 
-        if (float.TryParse(this.Temperature, out float temperature))
+        if (this.IsReasoningModel)
         {
-            request.Temperature = temperature;
+            this.MaxTokens = null;
+            this.Temperature = null; // property not supported for reasoning model.
         }
-
-        if (float.TryParse(this.TopP, out float topP))
+        else
         {
-            request.TopP = topP;
+            if (int.TryParse(this.MaxTokens, out int maxTokens))
+            {
+                request.MaxOutputTokenCount = maxTokens;
+            }
+
+            if (float.TryParse(this.Temperature, out float temperature))
+            {
+                request.Temperature = temperature;
+            }
         }
 
+        // ToDo: SetNewMaxCompletionTokensPropertyEnabled() has a bug in the current version 
+        // of the Azure.AI.OpenAI SDK but is fixed in the next preview.
+        // 
+        // This method doesn't swap max_tokens with max_completion_tokens and throws errors 
+        // if max_tokens is set. max_completion_tokens is not configurable due to this bug.
+        // 
+        // Hence, setting max_tokens to null for reasoning models until the fixed SDK version 
+        // can be adopted.
+        // request.SetNewMaxCompletionTokensPropertyEnabled(this.IsReasoningModel);
+
         return request;
     }
 }