Created by: lcy-seso

PaddlePaddle has a LoD Tensor, so it does not need to padding sequence in a batch. We are going to implement CRF as an operator in C++ so that users can easily use it regardless of processing any details of batching sequences in a batch.