In that case, I’d recommend you to separate them into different docker container and map to different port as @Jiang suggested - currently BentoML does not have GPU support, is that required for your model serving use case?