Round 2 Breaks Previous Code

With the release of round 2 and the OpenAI-friendly function calling interface, previous round 1 code is no longer working. I noticed that when using the previous provided code to parse function doc-strings, we occasionally get an error when trying to split on semicolon for parameter information. Specifically this line in the original provided code: agents/openai_agent.py · 115fc9e909ae7fc5052bdecf07bda587bd0aad95 · AIcrowd / Challenges / Sony Commonsense Persona-Grounded Dialogue Challenge 2025 / sony-cpdc-2025-starter-kit · GitLab

Is this going to be fixed? It’s a bit demoralizing to have to somewhat restart all the tuning done previous to round 2. Alternatively if the doc-string format in the private test set has changed, can we get some examples?

Hi @nicholas_liu ,

We are aware of the issue. We will take a look at why it happened and come up with appropriate solutions.

Hi @nicholas_liu ,

We have found the cause of the bug.

Basically, there are functions where no parameters are taken. In these cases, it will come in the following form.

    Parameters:
    ----------
    This function takes no parameters.

Thank you for the response