A machine learning team has a model that needs to serve predictions with very low latency (under 10 ms) for a real-time web application. The model is a small ensemble of three neural networks that fits in memory. Which SageMaker inference option is MOST appropriate?