-
Notifications
You must be signed in to change notification settings - Fork 10
Open
Description
Currently, the interface is synced. That is, for every input "x", the client will get some output "y=f(x)".
But for some application, the input/output may be asynced. Take speech recognition as an example, an typical usage may be,
- The client starts the recognition.
- The client keeps feeding audio data to the model, without receiving any output.
- When the model thinks the input audio data is enough to make a reasonable prediction, it will actively tells the client the result.
- The client stops the recognition.
So the client will need to provide an "OnPredictionResult" callback to the model.
In some cases, it is the model that actively asks for input (e.g. when the model thinks it is ready). Then there will be no Step 2 and the client needs to provide an "OnGetInput" callback to the model.
This is not a blocker for now, but just an interesting issue to think about.
Metadata
Metadata
Assignees
Labels
No labels