Update to M.E.AI 9.3.0-preview.1.25161.3 (#7414)

stephentoub · web-flow · commit 073516f2f63a · 2025-03-11T18:15:59.000-07:00
diff --git a/eng/Versions.props b/eng/Versions.props
@@ -46,7 +46,7 @@
     <MicrosoftDotNetInteractiveVersion>1.0.0-beta.24375.2</MicrosoftDotNetInteractiveVersion>
     <MicrosoftMLOnnxRuntimeVersion>1.18.1</MicrosoftMLOnnxRuntimeVersion>
     <MlNetMklDepsVersion>0.0.0.12</MlNetMklDepsVersion>
-    <MicrosoftExtensionsAIVersion>9.3.0-preview.1.25114.11</MicrosoftExtensionsAIVersion>
+    <MicrosoftExtensionsAIVersion>9.3.0-preview.1.25161.3</MicrosoftExtensionsAIVersion>
     <!-- runtime.native.System.Data.SqlClient.sni is not updated by dependency flow as it is not produced live anymore. -->
     <RuntimeNativeSystemDataSqlClientSniVersion>4.4.0</RuntimeNativeSystemDataSqlClientSniVersion>
     <!--
diff --git a/src/Microsoft.ML.GenAI.Core/CausalLMPipelineChatClient.cs b/src/Microsoft.ML.GenAI.Core/CausalLMPipelineChatClient.cs
@@ -33,9 +33,9 @@ public CausalLMPipelineChatClient(
         _pipeline = pipeline;
     }
 
-    public virtual Task<ChatResponse> GetResponseAsync(IList<ChatMessage> chatMessages, ChatOptions? options = null, CancellationToken cancellationToken = default)
+    public virtual Task<ChatResponse> GetResponseAsync(IEnumerable<ChatMessage> messages, ChatOptions? options = null, CancellationToken cancellationToken = default)
     {
-        var prompt = _chatTemplateBuilder.BuildPrompt(chatMessages, options);
+        var prompt = _chatTemplateBuilder.BuildPrompt(messages, options);
         var stopSequences = options?.StopSequences ?? Array.Empty<string>();
 
         var output = _pipeline.Generate(
@@ -49,30 +49,31 @@ public virtual Task<ChatResponse> GetResponseAsync(IList<ChatMessage> chatMessag
         {
             CreatedAt = DateTime.UtcNow,
             FinishReason = ChatFinishReason.Stop,
+            ResponseId = Guid.NewGuid().ToString("N"),
         });
     }
 
 #pragma warning disable CS1998 // Async method lacks 'await' operators and will run synchronously
     public virtual async IAsyncEnumerable<ChatResponseUpdate> GetStreamingResponseAsync(
 #pragma warning restore CS1998 // Async method lacks 'await' operators and will run synchronously
-        IList<ChatMessage> chatMessages,
+        IEnumerable<ChatMessage> messages,
         ChatOptions? options = null,
         [EnumeratorCancellation] CancellationToken cancellationToken = default)
     {
-        var prompt = _chatTemplateBuilder.BuildPrompt(chatMessages, options);
+        var prompt = _chatTemplateBuilder.BuildPrompt(messages, options);
         var stopSequences = options?.StopSequences ?? Array.Empty<string>();
 
+        string responseId = Guid.NewGuid().ToString("N");
         foreach (var output in _pipeline.GenerateStreaming(
             prompt,
             maxLen: options?.MaxOutputTokens ?? 1024,
             temperature: options?.Temperature ?? 0.7f,
             stopSequences: stopSequences.ToArray()))
         {
-            yield return new ChatResponseUpdate
+            yield return new(ChatRole.Assistant, output)
             {
-                Role = ChatRole.Assistant,
-                Text = output,
                 CreatedAt = DateTime.UtcNow,
+                ResponseId = responseId,
             };
         }
     }
diff --git a/src/Microsoft.ML.GenAI.Core/Utility/IChatTemplateBuilder.cs b/src/Microsoft.ML.GenAI.Core/Utility/IChatTemplateBuilder.cs
@@ -32,7 +32,7 @@ public interface IMEAIChatTemplateBuilder
     /// <param name="options"></param>
     /// <param name="appendAssistantTag">true if append assistant tag at the end of prompt.</param>
     /// <returns></returns>
-    string BuildPrompt(IList<ChatMessage> messages, ChatOptions? options = null, bool appendAssistantTag = true);
+    string BuildPrompt(IEnumerable<ChatMessage> messages, ChatOptions? options = null, bool appendAssistantTag = true);
 }
 
 public interface IChatTemplateBuilder : IAutoGenChatTemplateBuilder, ISemanticKernelChatTemplateBuilder
diff --git a/src/Microsoft.ML.GenAI.LLaMA/Llama3CausalLMChatClient.cs b/src/Microsoft.ML.GenAI.LLaMA/Llama3CausalLMChatClient.cs
@@ -25,7 +25,7 @@ public Llama3CausalLMChatClient(
     }
 
     public override Task<ChatResponse> GetResponseAsync(
-        IList<ChatMessage> chatMessages,
+        IEnumerable<ChatMessage> messages,
         ChatOptions? options = null,
         CancellationToken cancellationToken = default)
     {
@@ -40,18 +40,18 @@ public override Task<ChatResponse> GetResponseAsync(
             options.StopSequences = new List<string> { _eotToken };
         }
 
-        return base.GetResponseAsync(chatMessages, options, cancellationToken);
+        return base.GetResponseAsync(messages, options, cancellationToken);
     }
 
     public override IAsyncEnumerable<ChatResponseUpdate> GetStreamingResponseAsync(
-        IList<ChatMessage> chatMessages,
+        IEnumerable<ChatMessage> messages,
         ChatOptions? options = null,
         CancellationToken cancellationToken = default)
     {
         options ??= new ChatOptions();
         options.StopSequences ??= [];
         options.StopSequences.Add(_eotToken);
 
-        return base.GetStreamingResponseAsync(chatMessages, options, cancellationToken);
+        return base.GetStreamingResponseAsync(messages, options, cancellationToken);
     }
 }
diff --git a/src/Microsoft.ML.GenAI.LLaMA/Llama3_1ChatTemplateBuilder.cs b/src/Microsoft.ML.GenAI.LLaMA/Llama3_1ChatTemplateBuilder.cs
@@ -88,7 +88,7 @@ public string BuildPrompt(ChatHistory chatHistory)
         return sb.ToString();
     }
 
-    public string BuildPrompt(IList<ChatMessage> messages, ChatOptions? options = null, bool appendAssistantTag = true)
+    public string BuildPrompt(IEnumerable<ChatMessage> messages, ChatOptions? options = null, bool appendAssistantTag = true)
     {
         var availableRoles = new[] { ChatRole.System, ChatRole.User, ChatRole.Assistant };
         if (messages.Any(m => m.Text is null))
diff --git a/src/Microsoft.ML.GenAI.Phi/Phi3/Phi3CausalLMChatClient.cs b/src/Microsoft.ML.GenAI.Phi/Phi3/Phi3CausalLMChatClient.cs
@@ -30,7 +30,7 @@ public Phi3CausalLMChatClient(
     }
 
     public override Task<ChatResponse> GetResponseAsync(
-        IList<ChatMessage> chatMessages,
+        IEnumerable<ChatMessage> messages,
         ChatOptions? options = null,
         CancellationToken cancellationToken = default)
     {
@@ -45,18 +45,18 @@ public override Task<ChatResponse> GetResponseAsync(
             options.StopSequences = [_eotToken];
         }
 
-        return base.GetResponseAsync(chatMessages, options, cancellationToken);
+        return base.GetResponseAsync(messages, options, cancellationToken);
     }
 
     public override IAsyncEnumerable<ChatResponseUpdate> GetStreamingResponseAsync(
-        IList<ChatMessage> chatMessages,
+        IEnumerable<ChatMessage> messages,
         ChatOptions? options = null,
         CancellationToken cancellationToken = default)
     {
         options ??= new ChatOptions();
         options.StopSequences ??= [];
         options.StopSequences.Add(_eotToken);
 
-        return base.GetStreamingResponseAsync(chatMessages, options, cancellationToken);
+        return base.GetStreamingResponseAsync(messages, options, cancellationToken);
     }
 }
diff --git a/src/Microsoft.ML.GenAI.Phi/Phi3/Phi3ChatTemplateBuilder.cs b/src/Microsoft.ML.GenAI.Phi/Phi3/Phi3ChatTemplateBuilder.cs
@@ -89,7 +89,7 @@ public string BuildPrompt(ChatHistory chatHistory)
         return sb.ToString();
     }
 
-    public string BuildPrompt(IList<ChatMessage> messages, ChatOptions? options = null, bool appendAssistantTag = true)
+    public string BuildPrompt(IEnumerable<ChatMessage> messages, ChatOptions? options = null, bool appendAssistantTag = true)
     {
         var availableRoles = new[] { ChatRole.System, ChatRole.User, ChatRole.Assistant };
         if (messages.Any(m => m.Text is null))

Original file line number	Diff line number	Diff line change
`@@ -33,9 +33,9 @@ public CausalLMPipelineChatClient(`
`33`	`33`	`_pipeline = pipeline;`
`34`	`34`	`}`
`35`	`35`
`36`		`- public virtual Task<ChatResponse> GetResponseAsync(IList<ChatMessage> chatMessages, ChatOptions? options = null, CancellationToken cancellationToken = default)`
	`36`	`+ public virtual Task<ChatResponse> GetResponseAsync(IEnumerable<ChatMessage> messages, ChatOptions? options = null, CancellationToken cancellationToken = default)`
`37`	`37`	`{`
`38`		`- var prompt = _chatTemplateBuilder.BuildPrompt(chatMessages, options);`
	`38`	`+ var prompt = _chatTemplateBuilder.BuildPrompt(messages, options);`
`39`	`39`	`var stopSequences = options?.StopSequences ?? Array.Empty<string>();`
`40`	`40`
`41`	`41`	`var output = _pipeline.Generate(`
`@@ -49,30 +49,31 @@ public virtual Task<ChatResponse> GetResponseAsync(IList<ChatMessage> chatMessag`
`49`	`49`	`{`
`50`	`50`	`CreatedAt = DateTime.UtcNow,`
`51`	`51`	`FinishReason = ChatFinishReason.Stop,`
	`52`	`+ ResponseId = Guid.NewGuid().ToString("N"),`
`52`	`53`	`});`
`53`	`54`	`}`
`54`	`55`
`55`	`56`	`#pragma warning disable CS1998 // Async method lacks 'await' operators and will run synchronously`
`56`	`57`	`public virtual async IAsyncEnumerable<ChatResponseUpdate> GetStreamingResponseAsync(`
`57`	`58`	`#pragma warning restore CS1998 // Async method lacks 'await' operators and will run synchronously`
`58`		`- IList<ChatMessage> chatMessages,`
	`59`	`+ IEnumerable<ChatMessage> messages,`
`59`	`60`	`ChatOptions? options = null,`
`60`	`61`	`[EnumeratorCancellation] CancellationToken cancellationToken = default)`
`61`	`62`	`{`
`62`		`- var prompt = _chatTemplateBuilder.BuildPrompt(chatMessages, options);`
	`63`	`+ var prompt = _chatTemplateBuilder.BuildPrompt(messages, options);`
`63`	`64`	`var stopSequences = options?.StopSequences ?? Array.Empty<string>();`
`64`	`65`
	`66`	`+ string responseId = Guid.NewGuid().ToString("N");`
`65`	`67`	`foreach (var output in _pipeline.GenerateStreaming(`
`66`	`68`	`prompt,`
`67`	`69`	`maxLen: options?.MaxOutputTokens ?? 1024,`
`68`	`70`	`temperature: options?.Temperature ?? 0.7f,`
`69`	`71`	`stopSequences: stopSequences.ToArray()))`
`70`	`72`	`{`
`71`		`- yield return new ChatResponseUpdate`
	`73`	`+ yield return new(ChatRole.Assistant, output)`
`72`	`74`	`{`
`73`		`- Role = ChatRole.Assistant,`
`74`		`- Text = output,`
`75`	`75`	`CreatedAt = DateTime.UtcNow,`
	`76`	`+ ResponseId = responseId,`
`76`	`77`	`};`
`77`	`78`	`}`
`78`	`79`	`}`
Original file line number	Diff line number	Diff line change
`@@ -32,7 +32,7 @@ public interface IMEAIChatTemplateBuilder`
`32`	`32`	`/// <param name="options"></param>`
`33`	`33`	`/// <param name="appendAssistantTag">true if append assistant tag at the end of prompt.</param>`
`34`	`34`	`/// <returns></returns>`
`35`		`- string BuildPrompt(IList<ChatMessage> messages, ChatOptions? options = null, bool appendAssistantTag = true);`
	`35`	`+ string BuildPrompt(IEnumerable<ChatMessage> messages, ChatOptions? options = null, bool appendAssistantTag = true);`
`36`	`36`	`}`
`37`	`37`
`38`	`38`	`public interface IChatTemplateBuilder : IAutoGenChatTemplateBuilder, ISemanticKernelChatTemplateBuilder`
Original file line number	Diff line number	Diff line change
`@@ -25,7 +25,7 @@ public Llama3CausalLMChatClient(`
`25`	`25`	`}`
`26`	`26`
`27`	`27`	`public override Task<ChatResponse> GetResponseAsync(`
`28`		`- IList<ChatMessage> chatMessages,`
	`28`	`+ IEnumerable<ChatMessage> messages,`
`29`	`29`	`ChatOptions? options = null,`
`30`	`30`	`CancellationToken cancellationToken = default)`
`31`	`31`	`{`
`@@ -40,18 +40,18 @@ public override Task<ChatResponse> GetResponseAsync(`
`40`	`40`	`options.StopSequences = new List<string> { _eotToken };`
`41`	`41`	`}`
`42`	`42`
`43`		`- return base.GetResponseAsync(chatMessages, options, cancellationToken);`
	`43`	`+ return base.GetResponseAsync(messages, options, cancellationToken);`
`44`	`44`	`}`
`45`	`45`
`46`	`46`	`public override IAsyncEnumerable<ChatResponseUpdate> GetStreamingResponseAsync(`
`47`		`- IList<ChatMessage> chatMessages,`
	`47`	`+ IEnumerable<ChatMessage> messages,`
`48`	`48`	`ChatOptions? options = null,`
`49`	`49`	`CancellationToken cancellationToken = default)`
`50`	`50`	`{`
`51`	`51`	`options ??= new ChatOptions();`
`52`	`52`	`options.StopSequences ??= [];`
`53`	`53`	`options.StopSequences.Add(_eotToken);`
`54`	`54`
`55`		`- return base.GetStreamingResponseAsync(chatMessages, options, cancellationToken);`
	`55`	`+ return base.GetStreamingResponseAsync(messages, options, cancellationToken);`
`56`	`56`	`}`
`57`	`57`	`}`
Original file line number	Diff line number	Diff line change
`@@ -88,7 +88,7 @@ public string BuildPrompt(ChatHistory chatHistory)`
`88`	`88`	`return sb.ToString();`
`89`	`89`	`}`
`90`	`90`
`91`		`- public string BuildPrompt(IList<ChatMessage> messages, ChatOptions? options = null, bool appendAssistantTag = true)`
	`91`	`+ public string BuildPrompt(IEnumerable<ChatMessage> messages, ChatOptions? options = null, bool appendAssistantTag = true)`
`92`	`92`	`{`
`93`	`93`	`var availableRoles = new[] { ChatRole.System, ChatRole.User, ChatRole.Assistant };`
`94`	`94`	`if (messages.Any(m => m.Text is null))`
Original file line number	Diff line number	Diff line change
`@@ -30,7 +30,7 @@ public Phi3CausalLMChatClient(`
`30`	`30`	`}`
`31`	`31`
`32`	`32`	`public override Task<ChatResponse> GetResponseAsync(`
`33`		`- IList<ChatMessage> chatMessages,`
	`33`	`+ IEnumerable<ChatMessage> messages,`
`34`	`34`	`ChatOptions? options = null,`
`35`	`35`	`CancellationToken cancellationToken = default)`
`36`	`36`	`{`
`@@ -45,18 +45,18 @@ public override Task<ChatResponse> GetResponseAsync(`
`45`	`45`	`options.StopSequences = [_eotToken];`
`46`	`46`	`}`
`47`	`47`
`48`		`- return base.GetResponseAsync(chatMessages, options, cancellationToken);`
	`48`	`+ return base.GetResponseAsync(messages, options, cancellationToken);`
`49`	`49`	`}`
`50`	`50`
`51`	`51`	`public override IAsyncEnumerable<ChatResponseUpdate> GetStreamingResponseAsync(`
`52`		`- IList<ChatMessage> chatMessages,`
	`52`	`+ IEnumerable<ChatMessage> messages,`
`53`	`53`	`ChatOptions? options = null,`
`54`	`54`	`CancellationToken cancellationToken = default)`
`55`	`55`	`{`
`56`	`56`	`options ??= new ChatOptions();`
`57`	`57`	`options.StopSequences ??= [];`
`58`	`58`	`options.StopSequences.Add(_eotToken);`
`59`	`59`
`60`		`- return base.GetStreamingResponseAsync(chatMessages, options, cancellationToken);`
	`60`	`+ return base.GetStreamingResponseAsync(messages, options, cancellationToken);`
`61`	`61`	`}`
`62`	`62`	`}`
Original file line number	Diff line number	Diff line change
`@@ -89,7 +89,7 @@ public string BuildPrompt(ChatHistory chatHistory)`
`89`	`89`	`return sb.ToString();`
`90`	`90`	`}`
`91`	`91`
`92`		`- public string BuildPrompt(IList<ChatMessage> messages, ChatOptions? options = null, bool appendAssistantTag = true)`
	`92`	`+ public string BuildPrompt(IEnumerable<ChatMessage> messages, ChatOptions? options = null, bool appendAssistantTag = true)`
`93`	`93`	`{`
`94`	`94`	`var availableRoles = new[] { ChatRole.System, ChatRole.User, ChatRole.Assistant };`
`95`	`95`	`if (messages.Any(m => m.Text is null))`