* add -linear-schedule-with-warmup

* update lr scheduler usage in run_glue

* add import info

* add other 5 scheduler with warmup

* add copyright, update usage in run_glue

* simplify argument, make warmup arg support float and int

* Add some LambdaDecay scheduler with warmup, update usage in run_glue

* update classname and unify two cosine decays into one

* update usage in run_glue

* fix typo

* update WarmUp to Warmup, and update class name about Const, and update doc, and usage

* update usage of decay class

* update usage of decay class

* add seq2seq infer

* update argument description, remove useless import

* add deploy directory

* add deploy directory, add relative import

* update arg usage in README, fix import order.

* dy2sta and inference support

* delete useless code

* update dir

* sys

* delete useless import

* update docs

* update codes

* update docs

* update codes

* update codes

* update codes

* add method tokenizer.get_special_tokens_mask()

* Add inference support for BERT.

* Update inference support for BERT.

* fix checkpoints problem.

* bert benchmark fleet support pure_fp16

* fix use_pure_fp16

add an optional boolean argument pad_rght to Pad class to support padding direction selection

fix docs for visualdl

* move coupletdataset to paddlenlp.dataset

* move coupletdataset to paddlenlp.dataset

* add doc for coupletdataset copyright, fix other api's sample code

* fix reader

* increase epoch from 100 to 200

* move coupletdataset to paddlenlp.dataset

* move coupletdataset to paddlenlp.dataset

* update docs

* add imdb dataset

* update codes

export text classification model in static graph 

fix crf bug in paddlenlp

fix the pure fp16 of argument (#5236)

add gpt2 model for the paddlenlp

* refine transformer cpu

* delete run_pretrain.sh

* update docs