fix oom for batch_size=208

* Enhance dataloader robustness

* Fix the bug about unexpected exiting when reading image failed

* new usage of amp training.

* change the usage of amp and pure fp16 training.

* modified code as reviews

* fix vgg stop grad

* beautify code

* refine distillation model

* fix vgg stop grad

* beautify code

* Update xception.py

* Update vgg.py

* Update regnet.py

* Update distillation_models.py

* Update res2net.py

* Update res2net_vd.py

* Update resnest.py

* add 1230 FAQ

* modify according to comments

* change FAQ update date in readme

* Adapt to PaddleHub2.0 to eliminate warning
* Fix the training log format

* fix predict

* fix export model

* fix doc

* Fix the timing of hubserving

* fix reshape to flatten

* fix reshape

This reverts commit d5105825.

* add amp support for SE_ResNet50_vd

* updates for data_format

* add pretrained models

* fix typo

* add enable benchmark for lite

* fix config

* Updata "*" to "paddle.multiply"

* Update "+" to "paddle.add"

* Fix "tar" to "pdparams" to adapt to dygraph

* Update the download link of Paddle Inference Library