Abstract: Performing training and inference for Graph Neural Networks (GNNs) under tight latency constraints has become increasingly difficult as real-world input graphs continue to grow. Compared to ...