Hi,
I found in the official tutorial that the method of compiling onnx and compiling mxnet models are different, where mxnet used relay.build
, but onnx used relay.build_module
.
Does these two methods do the same thing?
The example in the tutorial only shows how to dump the compiled model built from relay.build
, could I dump and reload the compiled model built from relay.build_module.create_executor
?
By the way, I am also confused that - do I have to take care of the train
mode and eval
mode of some operators such as batch norm
or dropout
?