[Bug]: 'name' field on messages not working when using Mistral.AI's API

[marklysze]

Describe the bug

Mistral.AI's API is strict on what keys are allowed on the messages.

When running a GroupChat, the name key on each message is causing an exception when using the Mistral.AI API through the standard client:

openai.UnprocessableEntityError: Error code: 422 - {'object': 'error', 'message': {'detail': [{'type': 'extra_forbidden', 'loc': ['body', 'messages', 1, 'user', 'name'], 'msg': 'Extra inputs are not permitted', 'input': 'User_proxy'}]}, 'type': 'invalid_request_error', 'param': None, 'code': None}

If the name key is removed from each message no exception is thrown.

Mistral.AI API's documentation on message:

Creating this issue in relation to testing from PR #2635, specific messages 1 and 2

---

Would be good to have an endpoint that developers could use for Mistral.AI, and other non-OpenAI inference, where the name isn't accepted/used.

On a side-note, my testing shows that the name field isn't being used for Ollama, LiteLLM+Ollama, or together.ai.

Steps to reproduce

No response

Model Used

No response

Expected Behavior

No response

Screenshots and logs

No response

Additional Information

No response

[ekzhu]

A possible solution is to have a built-in model client for Mistral AI's API. It's a thin client that does a simple pop of the name key. See example of model client: https://microsoft.github.io/autogen/docs/topics/non-openai-models/cloud-anthropic

[marklysze]

A possible solution is to have a built-in model client for Mistral AI's API. It's a thin client that does a simple pop of the name key. See example of model client: https://microsoft.github.io/autogen/docs/topics/non-openai-models/cloud-anthropic

Thanks @ekzhu, for Mistral, they have their own library, like the Anthropic one, so would you see this using that library or the openai library?

from mistralai.client import MistralClient
from mistralai.models.chat_completion import ChatMessage

[ekzhu]

I think using their library as an optional dependency makes sense. So when api_type=mistral_ai, it is going to check if the mistral client library has been installed if not then raise runtime error.

[marklysze]

I think using their library as an optional dependency makes sense. So when api_type=mistral_ai, it is going to check if the mistral client library has been installed if not then raise runtime error.

Okay, I've started creating a Mistral client model class.

[marklysze]

Hey @ekzhu, I've created a Mistral Client class. This class does cater for the removal of the name field. PR #2892.

[marklysze]

Yay, Mistral client class merged - please use api_type='mistral' in your configs to use it :).