* Place mixed precision inside PaddleDetection

roll back to the monkey patch version as a temporary measure, before it is
merged into paddle

* Add command flag for `loss_scale`

* Fix a stupid indentation error

optimizer should be in the mixed precision context

* Initial FP16 training

* Add mixed precision training to rest of the detection models

* Revert "Add support for mixed precision training (#3406)"

This reverts commit 3a2c1062.

* Bug fixes and some tweaks

* fix_bug_server

* fix_readme

* Support finetune by custom dataset

* add finetune args

* add load finetune

* reconstruct load

* add transfer learning doc

* add fruit demo

* add quick start

* add data preprocessing FAQ

* using new load api

* merge

* add dygraph user guide in models/dygraph

* remove useless files

* fix deepfm failed in 2*1 local distributed training

* fix deepfm preprocess bug

* fix deepfm preprocess bug and README

test=develop

* fix quantization doc 

Update PaddleSlim LightNAS

* update

* update

* update

* fix run_ernie.sh

* Update README.md

fix path of dir of files in windows

* PaddleRec xdeepfm add local distributed training

* PaddleRec deepfm dcn add local distributed training

* change PaddleRec deepfm README.md

* add embedding is_sparse flag in argparser

* fix deepfm failed in ver 1.5.2 when use transpiler

* fix xdeepfm failed in ver 1.5.2 when use transpiler

* fix some bugs in dnn deepfm and dcn

* fix embedding size in deepfm

* MRQA2019-D-NET

* delete_data_file

* add distillation demo and doc

* add classification quantization demo

* Disable optimizer fusion for RCNN models

waiting for upstream fix

* Disable completely

* Add support for mixed precision training

* Disable `fuse_all_reduce_ops` when training in mixed precision

waiting for upstream fix

* Add pruning demo for classification

* set encoding=utf8 for open

* set encoding=utf8 for open

* Initial support for distributed training

* Housekeep on rank 0 only

* Conform to models convention

add license for nlp models

* fix model save

* fix doc

* update TensorRT 1.5.2 inference time

* add EfficientNet configuration in README

* add notes for the ported weight of EfficientNet

* fix_infer_pix2pixreader

Add EfficientNet

Fix transformer inference when using ParallelExecutor by setting ExecutionStrategy to clear tensor array after each iteration. (#3374)