Yeah, should be 'elastic' and 'very reduced runtime, possibly just inferencing', so exposing Triton to an API gateway and putting a custom load balancer and tas

Yeah, should be 'elastic' and 'very reduced runtime, possibly just inferencing', so exposing Triton to an API gateway and putting a custom load balancer and task queue facade ? Curious too. Also, GPUs do other things than inferencing, right?