Convert_messages in Version 1.5.2

[ReneFr81]

Hello all,

First of all, I want to thank you for developing this library! I'm quite new to using instructor and appreciate the thoughtful design behind it.

I'm currently working on tool calling and noticed a change in behavior starting from version 1.5.2, specifically after the multimodal support was added. Previously, I was able to pass tool responses back to the LLM as structured tool messages like the example below:

[
  {"role": "system", "content": "You are an AI assistant..."},
  {"role": "user", "content": "user prompt"},
  {
    "role": "assistant", "content": null, "tool_calls": [
      {"id": "a4de1832-71bb-483d-9b58-85b97c11639e", "function": {"name": "Search...", "arguments": "{\"a\": \"1234\"}"}, "type": "function"}
    ]
  },
  {"role": "tool", "content": "{...}", "tool_call_id": "a4de1832-71bb-483d-9b58-85b97c11639e"}
]

Since convert_messages now only retains role and content fields, other custom attributes like tool_call_id are not included, causing issues in workflows that rely on structured tool messages.

To address this, I modified the convert_messages function as follows:

def convert_messages(
    messages: list[dict[str, Union[str, list[Union[str, Image]]]]],
    mode: Mode,
) -> list[dict[str, Any]]:
    converted_messages = []
    for message in messages:
        role = message["role"]
        content = message["content"]
        if isinstance(content, str):
            converted_messages.append(message)  # Pass the entire message to retain all fields
        else:
            converted_content = convert_contents(content, mode)
            converted_messages.append({"role": role, "content": converted_content})
    return converted_messages

This change ensures that all fields in the original message, like tool_call_id, are preserved. I wanted to ask if this approach is appropriate or if there’s a better way to handle this scenario? Is there a reason convert_messages only processes role and content fields?

Any guidance or recommendations would be greatly appreciated!

Thank you!