Explain how to generate module library in Quick Start tutorial (#323)

* Explain how to generate module library

* Small fix

nnvm/tutorials/define_and_compile_model.py

@@ -68,6 +68,9 @@ print(net.debug_str())
 # in an optimized runtime module for model serving.
 #
 # We'll first compile for Nvidia GPU.
+# To generate the module library, TVM will first transfer graph IR into lower
+# intrinsic IR for the specified target backend, which is CUDA in this example.
+# Then target backend will generate module library.
 
 opt_level = 0
 target = tvm.target.cuda()