text-generation-inference

Changing the waiting_served_ratio default (stack more aggressively by default).

#1820

Merged