description DNN inference scheduling framework to improve GPU utilization under SLO constraints. Serving Heterogeneous Machine Learning Models on Multi-GPU Servers with Spatio-Temporal Sharing