.Net: KernelFunctionSelectionStrategy.SelectAgentAsync and KernelFunctionTerminationStrategy.ShouldAgentTerminateAsync does not support streaming.

[0xLiao]

While following the documentation to use the Agent Collaboration capability, since the model only supports streaming mode, I modified

await foreach (ChatMessageContent response in chat.InvokeAsync())

to

await foreach (StreamingChatMessageContent response in chat.InvokeStreamingAsync())

The first Agent sent the request with streaming enabled, send:

"model":"Qwen/QwQ-32B","stream":true,"stream_options":{"include_usage":true}"

but immediately threw an exception afterward.

{"error":{"code":"invalid_parameter_error","param":null,"message":"This model only support stream mode, please enable the stream parameter to access the model. ","type":"invalid_request_error"}
      Microsoft.SemanticKernel.HttpOperationException: HTTP 400 (: 10137)
System.AggregateException: One or more errors occurred. (HTTP 400 (: 10137
)

After analyzing, I found that while the first request uses streaming mode, it switches to non-streaming when .SelectAgentAsync selects the next Agent：

at Microsoft.SemanticKernel.Connectors.OpenAI.ClientCore.RunRequestAsync[T](Func`1 request)
   at Microsoft.SemanticKernel.Connectors.OpenAI.ClientCore.GetChatMessageContentsAsync(String targetModel, ChatHistory chatHistory, PromptExecutionSettings executionSettings, Kernel kernel, CancellationToken cancellationToken)
   at Microsoft.SemanticKernel.KernelFunctionFromPrompt.GetChatCompletionResultAsync(IChatCompletionService chatCompletion, Kernel kernel, PromptRenderingResult promptRenderingResult, CancellationToken cancellationToken)
   at Microsoft.SemanticKernel.KernelFunctionFromPrompt.InvokeCoreAsync(Kernel kernel, KernelArguments arguments, CancellationToken cancellationToken)
   at Microsoft.SemanticKernel.KernelFunction.<>c__DisplayClass27_0.<<InvokeAsync>b__0>d.MoveNext()
--- End of stack trace from previous location ---
   at Microsoft.SemanticKernel.Kernel.InvokeFilterOrFunctionAsync(NonNullCollection`1 functionFilters, Func`2 functionCallback, FunctionInvocationContext context, Int32 index)
   at Microsoft.SemanticKernel.Kernel.OnFunctionInvocationAsync(KernelFunction function, KernelArguments arguments, FunctionResult functionResult, Boolean isStreaming, Func`2 functionCallback, CancellationToken cancellationToken)
   at Microsoft.SemanticKernel.KernelFunction.InvokeAsync(Kernel kernel, KernelArguments arguments, CancellationToken cancellationToken)
   at Microsoft.SemanticKernel.Agents.Chat.KernelFunctionSelectionStrategy.SelectAgentAsync(IReadOnlyList`1 agents, IReadOnlyList`1 history, CancellationToken cancellationToken)
   at Microsoft.SemanticKernel.Agents.Chat.SelectionStrategy.NextAsync(IReadOnlyList`1 agents, IReadOnlyList`1 history, CancellationToken cancellationToken)
   at Microsoft.SemanticKernel.Agents.AgentGroupChat.SelectAgentAsync(CancellationToken cancellationToken)
   at Microsoft.SemanticKernel.Agents.AgentGroupChat.InvokeStreamingAsync(CancellationToken cancellationToken)+MoveNext()

Although I explicitly used AgentGroupChat.InvokeStreamingAsync, the KernelFunctionSelectionStrategy.SelectAgentAsync method always calls KernelFunction.InvokeAsync internally. I found in the source code that it's hardcoded to use non-streaming invocation without any streaming support:

    protected sealed override async Task<Agent> SelectAgentAsync(IReadOnlyList<Agent> agents, IReadOnlyList<ChatMessageContent> history, CancellationToken cancellationToken = default)
    {
        history = await history.ReduceAsync(this.HistoryReducer, cancellationToken).ConfigureAwait(false);

        KernelArguments originalArguments = this.Arguments ?? [];
        KernelArguments arguments =
            new(originalArguments, originalArguments.ExecutionSettings?.ToDictionary(kvp => kvp.Key, kvp => kvp.Value))
            {
                { this.AgentsVariableName, string.Join(",", agents.Select(a => a.Name)) },
                { this.HistoryVariableName, ChatMessageForPrompt.Format(history, this.EvaluateNameOnly) },
            };

        this.Logger.LogKernelFunctionSelectionStrategyInvokingFunction(nameof(NextAsync), this.Function.PluginName, this.Function.Name);

        FunctionResult result = await this.Function.InvokeAsync(this.Kernel, arguments, cancellationToken).ConfigureAwait(false);

It appears there's no external configuration option to force streaming mode behavior in this case.

Platform

• .NET
• Microsoft.SemanticKernel 1.45.0
• model: Qwen/QwQ-32B
• Windows, Mac