* 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 eval script

* fix dali shell

* Add FAQ 20201207

* Fix FAQ20201207

add se hrnet64_c

* fix pred

* fix resnest

* fix hrnet se

* fix se for export

* add shells, test=develop

fix resnest, ghostnet typo

* fix download link

* fix dpn

* fix se resnext101 metrics

* Update mobilenet_v3.py

* Update densenet.py

* Update resnest.py

* Update hrnet.py

* Update vgg.py

* Update xception.py

add GhostNet_x1_3_ssld_pretrained

* add support for multi cards eva

* Modify the "cls_model_dir" to "cls_model_path" and "cls_params_path"