* refine tool dir

* add support for infer

* add export model support

* fix yaml

* fix post process name

* fix topk name

* add support for train and eval

* rm unsed code

* add support for metric save and load ckp

* Enable profiler for static training.

* Polish the initialize of ProfilerOptions.

* Enable profiler for dynamic mode.

* adapt to net.eval for the framework just contains training flag setting
* fix bug when export swin transformer

* add multi xpu support for PaddleClas about dygraph

* add dygraph multi xpu support

* fix repvgg eval

* fix dp training

* fix single card train

* fix loss reduce from dict to list

* remove note

* add find_unused_parameters param

* fix default val

* support bs>1

* Support Visual DL

* Fix VDL

* Add doc of VDL, test=document_fix

* Add the en doc of VDL, test=document_fix
Co-authored-by: Nlittletomatodonkey <2120160898@bit.edu.cn>

* new usage of amp training.

* change the usage of amp and pure fp16 training.

* modified code as reviews

* add eta info

* rm duplicate desc

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

* fix predict

* fix export model

* fix doc