A few questions about compiling the DL model


I reading the official tutorial recently about Compile DL Models,such as TFlite, Compile the model with relay,the following code

with relay.build_module.build_config(opt_level=3):
graph, lib, params = relay.build(func, target, params=params

Does the optimization here refer only to the graph optimization performed by Relay?If so, where is the internal optimization for TVM?Such as optimizing tensor operators,if not,why the next step is Execute on TVM