© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Why do we need graph dispatcher 1. Minimal overhead: only one dispatching operation is required for each inference. 2. Fit for tvm.relay.transform.dispatch_global_func(mod, "main", {input_name: input_shape}, tvm.relay.vm.exp_dispatcher) vmc = relay.backend.vm.VMCompiler() with tvm.autotvm.apply_graph_best("resnet50_v1_graph_opt