Created by: guru4elephant

please public prediction benchmark with predict() and batch_predict() interface. For imdb task, the model should run on cpu device and model inference time should be profiled.

https://github.com/PaddlePaddle/Serving/pull/215