Autogen handoff

[tinased95]

Hi, I’m using the Handoff design pattern example, and it works well overall. However, when I ask a sales-related question, the task is successfully delegated to the SalesAgent, and the conversation proceeds smoothly. But when I ask something irrelevant to SalesAgent, it says it’s delegating back to triage, then fails with this error:

openai.InternalServerError: Error code: 500 - {'error': {'message': 'The model produced invalid content. Consider modifying your prompt if you are seeing this error persistently.', 'type': 'model_error', 'param': None, 'code': None}}

In short, triage can delegate tasks to the sub agents but when sub agents want to delegate task to triage again it fails

[ekzhu]

This is a 500 internal sever error we have started to see recently from open ai endpoints. Which model are you using?

It's not something we can fix at this point as it comes from open ai.

[tinased95]

The only change I added to the notebook is to use AzureOpenAIChatCompletionClient instead of OpenAIChatCompletionClient

[tinased95]

I'm using this setting for AzureOpenAIChatCompletionClient

model_client = AzureOpenAIChatCompletionClient(
    model="test-gpt4o",
    api_version="2023-03-15-preview",
    azure_endpoint=azure_endpoint,
    azure_ad_token_provider=token_provider,
    model_capabilities={
        "vision":False,
        "function_calling":True,
        "json_output":True,
    }
)

[JComyn]

Hi, did you find any solution to this problem? I'm facing a very similar error and also using AzureOpenAIChatCompletionClient.

[ekzhu]

Can you use a more recent gpt-4o version?

[JComyn]

Using other models, like gpt-4, solved the issue.