* add op_function_generator.exe retry in windows, test=develop

* modify test_imperative_using_non_zero_gpu from use two gpus to one gpu, test=develop

* remove objgraph

* remove graphviz

* fix ut

* add adaptivelsgd

* Todo fix the code to avoid the conflict.

Error description optimize for the math dir

fix judge cache file of inference api more accurate

test=develop

* support load state_dict from save_params/persistables

* remove failed unittest

* add load eof check & unittest

* remove eof check

* fix error message

* polish framework error message part 7

* fix typo

* polish by reviewes comment

* add input_spec & output_spec for translated_layer

* update error message

update the error message check for the some ops

add the error message check for the some operator

* fix ports conflict when launching multi-nodes in paddlecloud;test=develop

* add DISTRIBUTED_TRAINER_ENDPOINTS env for cloud;test=develop

* fix strategy, test=develop

* fix can_apply

* optimize slice TRT plugin

This patch removes unnecessary barrier for data transfer of needed offset,
so data transfer can be overlap with GPU kernel execution.

This patch also fixes incorrect name of slice plugin. That is, replaces
"layernorm" with "slice"

test=develop

* add serialize/deserialize to slice plugin

* add static shape slice trt plugin

* fix slice trt op convertor dynamic shape bug

* fix format by clang-format

* fix pylint format error

* fix problems commented by peiyang
Co-authored-by: NRyan Jeng <rjeng@nvidia.com>

Add *.bat file for building compile environment on windows

* polish framework error msg part 6

* polish lossed item

* fix failed unittest

* polish by reviewer comments

* optimize errror report

* add test case for pad op converter

* fix some spelling mistake commented by peiyang

* replace sequence length attr to input

* Remove the cache in post_traning_quantization, test=develop

* fix trt_dynamic_shape_ernie_deserialize_test

* support when opt cache dir does not exist