Mistral model client integration

[fm1320]

Description & Motivation

As part of extending Adalflow's capability we want to integrate Mistral as a provider

Pitch

No response

Provider POC

No response

Additional context

No response

[adjahossoualexandre]

Proposal

I will wrap Mistral text generation and embedding endpoints into a concrete implementation of the 6 ModelClient main methods.

I will use Mistal SDK.

from adalflow.core.model_client import ModelClient

def MistralClient(ModelClient):

    def __init__(self, model_name: Optional[str] = None) -> None:
        super().__init__()
        self._model_name = model_name

    def init_sync_client(self):
        pass

    def convert_inputs_to_api_kwargs(
        self,
        input: Any,  # for retriever, it is a single query,
        model_kwargs: dict = {},
        model_type: ModelType = ModelType.UNDEFINED,
    ) -> dict:
        pass

    def parse_embedding_response(self, response: Any) -> EmbedderOutput:
        pass

    def parse_chat_completion(self, completion: Any) -> GeneratorOutput:
        pass

    def call(self, api_kwargs: Dict = {}, model_type: ModelType = ModelType.UNDEFINED):
        pass

Regarding the init_async_client() method, I'm not sure how to implement it since the SDK has no dedicated method for it. Any suggestions?

[fm1320]

@adjahossoualexandre So in adalflow/adalflow/components/model_client basically you would create a new model client .py file similar to the example of the openai or other clients you see there

[adjahossoualexandre]

@fm1320 Yes, exactly. The only thing I'm not sure about is the init_async_client() because the SDK has no dedicated method for it but instead uses async def. Do you have any suggestions?

[fm1320]

Maybe wrap the synchronous client that makes the API call for async usage ? This is my rough idea, feel free to start the PR and I will adapt this part if needed

async def acall(self, api_kwargs: Dict = {}):
        response = await asyncio.to_thread(self.call, api_kwargs)
        return response
        
        
# Using the client
async def example():
    client = MistralClient()
    

    response = await client.acall(
        {"prompt": "Hello!"},
        model_type=ModelType.LLM
    )
    return response

[adjahossoualexandre]

I'll try to do that thanks!