explain the lowering process in nnvm.compiler.build (#339)

nnvm/python/nnvm/compiler/build_module.py

@@ -207,7 +207,7 @@ def build(graph, target=None, shape=None, dtype="float32", params=None, target_h
         The final execution graph.
 
     libmod : tvm.Module
-        The modue that comes with the execution graph
+        The module that comes with the execution graph
 
     params : dict of str to NDArray
         The updated parameters of graph if params is passed.
@@ -236,7 +236,7 @@ def build(graph, target=None, shape=None, dtype="float32", params=None, target_h
     if params and cfg.pass_enabled("PrecomputePrune"):
         graph, params = precompute_prune(graph, params)
         shape, dtype = _update_shape_dtype(shape, dtype, params)
-    # Operator Fusion and generatiom
+    # Operator Fusion and generation
     graph = graph_attr.set_shape_inputs(graph, shape)
     graph = graph_attr.set_dtype_inputs(graph, dtype)
     graph._set_json_attr("target", str(target), "str")

nnvm/tutorials/define_and_compile_model.py

@@ -67,10 +67,13 @@ print(net.debug_str())
 # optimization while TVM does the tensor-level optimization, resulting
 # 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.
+# We'll first compile for Nvidia GPU. Behind the scene, `nnvm.compiler.build`
+# first does a number of graph-level optimizations, e.g. pruning, fusing, etc.,
+# then registers the operators (i.e. the nodes of the optmized graphs) to 
+# TVM implementations to generate a `tvm.module`.
+# To generate the module library, TVM will first transfer the HLO IR into the lower
+# intrinsic IR of the specified target backend, which is CUDA in this example.
+# Then the machine code will be generated as the module library.
 
 opt_level = 0
 target = tvm.target.cuda()