How to enable GPT-4's JSON mode for Semantic Kernel?

[wu-hongjun]

Hi Everyone,

I want to ask a simple question, how to enable GPT4's JSON mode using SK?

I tried the following and it doesn't work:

# Add the OpenAI service
kernel.add_service(
    OpenAIChatCompletion(
        service_id="default",
        ai_model_id="gpt-4o",
        api_key=api_key,
        default_headers={
            "Content-Type": "application/json",
            "response_format": json.dumps({"type": "json_object"}),
        },
    )

[wu-hongjun]

Nevermind, I figured it out. Putting it here in case other people have similar problems:

You will need to specify this in the config.json:

"execution_settings": {
    "default": {
      "max_tokens": 800,
      "temperature": 0.5,
      "top_p": 1.0,
      "presence_penalty": 0.0,
      "frequency_penalty": 0.0,
      "response_format": { "type": "json_object" }
    }
  },