add batching in ask response stream

[yujonglee]

currently we are using model that is "fast enough", so not needed at the moment