* new usage of amp training.

* change the usage of amp and pure fp16 training.

* modified code as reviews

* Add RepVGG models

* Update model code

* Rename code

* format code

* add README of RepVGG

* update configs of ViT and DeiT

* update readme of DeiT models

* update readme of DeiT models

* update readme of DeiT models

* update README of ViT models

* Update the discription about ViT and DeiT

* Fix the error file

* Unified data format

* Add the DeiT models

* fix a training bug

* update the model init config

* update the License

* Add the ViT model

This reverts commit d5105825.

* add amp support for SE_ResNet50_vd

* updates for data_format

Fix some description of documentations

* support fp16 training

* Use compiled training program

* Change timing ips.

* Use dali

* add pure fp16 training

* fix a bug, which will not use fuse pass using pure fp16 training.

* modify code as review

* modify loss, so that it will use different loss when using pure fp16 training.

* remove some fluid API

* add static optimizer.

fix resnest, ghostnet typo

add resnest101

* unify model and model_name

* rm unsed archs fix bug of se_resnext

* fix distillation model for distributed.launch

* fix export model for googlenet

* Add InceptionV3

polish codes and docs to adapt Paddle2.0-rc