Abstract: Current deep learning compilers have made significant strides in optimizing computation graphs for single- and multi-model scenarios. However, they lack specific optimizations for ...