No function call found in the response

[happyhohoh]

I'm trying to replicate the Gobang experiment, but I'm consistently running into an issue where the LLM (including GPT-4o) does not return a function_call object. Instead, it returns natural language and markdown code blocks in the content field, which causes a No function call found error.

[wty500]

Hi,
We just re-ran our code using both GPT-4o and GPT-4.1, and in each case the model returned the correct function_call object:

Please try the following:

• Ensure you’re using the official OpenAI API endpoint that supports function calls.
• Re-run the experiment.

Let me know if you continue to run into issues.

[happyhohoh]

Thank you for your reply, ! I can't use official OpenAI API, but I use the OpenAI API from other websites. Maybe this is a problem with the website?I also tried DeepSeek API at the same time, and it produced the same issue.

[wty500]

Hi,

It is likely that the issue is caused by the third-party API you’re using. Some unofficial APIs may not fully align with the behavior of the official OpenAI API, especially in how function calls are handled.

We were able to reproduce the same issue with the DeepSeek API. It did not return a function call response, even when our prompt explicitly required one. This might be due to differences in the function calling format. DeepSeek’s format (as documented [here](https://api-docs.deepseek.com/guides/function_calling)) doesn’t fully align with the one defined in our llm.py.

If you still prefer to use DeepSeek, you could try adjusting the tools variable in llm.py to match the format expected by DeepSeek’s API. For instance, adding something like "type": "function","function":{...} in the tools variable.

[happyhohoh]

Sorry to ask you again, I debugged the api, the argument in the function_call in the returned result will be undefined, this should be a problem with json parsing, but I have not found a good way to solve it, which will cause the agent to not recognize correctly, I would like to know if this situation will occur when you call the api, thank you very much for reading my question

[wty500]

Thank you for your feedback. Here we post a valid response from the official GPT-4.1 api.

It seems that you should change content from "" to None