Question : Asynchronous capabilities for mlserver mlflow runtime

[ylambruschi]

It seems that mlflow runtime only handle predict methods of its model in a synchronous way when called from

MLServer/runtimes/mlflow/mlserver_mlflow/runtime.py

Line 202 in 3c9b5dd

model_output = self._model.predict(decoded_payload, params=params)

MLServer/runtimes/mlflow/mlserver_mlflow/runtime.py

Line 140 in 3c9b5dd

raw_predictions = self._model.predict(data, params=inference_params)

I know mlflow embedded serve() uses Flask and is not designed to be async, but there is no hard interface contract expecting predict() to be synchronous right ?
Let's say i develop my model and set the predict() method to be async. would it make sense to work on the possibility to await it in mlserver mlflow runtime to make all the chain async ? ( maybe with some option)