* fix sample code diff for OP(retinanet_detection_output, retinanet_target_assign), fix  default value for OP(sigmoid_focal_loss). test=develop

* fix float value. test=develop

* add int type for gamma and alpha of sigmoid_focal_loss.test=develop

fix Chinese doc bug in argsort, add the dtype support of dtypes

fix sample code diff for OP(retinanet_detection_output, retinanet_target_assign), fix `gamma` default value for OP(sigmoid_focal_loss) (#1971)

* fix sample code diff for OP(retinanet_detection_output, retinanet_target_assign), fix  default value for OP(sigmoid_focal_loss). test=develop

* fix float value. test=develop

Refine the doc of create_global_var, pad, pad_constant_like and pad2d to make it consistent with en doc.

* add inplace_abn. test=develop

* fix lstm doc error

* fix lstm doc error

* remove unused line in lstm doc

* Add learning_rate in NoamDecay test=develop

* Update sample code test=develop

* add update for gelu: approximation option

* removefill_constant_batch_size_like  gaussian_random_batch_size_like  uniform_random_batch_size_like_cn API doc test=develop

* add filter_by_instag out_val_ifempty cn api test=develop

* change out_val_ifempty to out_val_if_empty in filter_by_instag test=develop

In elementwise_min api doc has the error, all the api just use the elementwise_max op, we just correct the doc.

Fix the pow doc bug, change the pow parameter from 2.0 to 3.0

Fix the compare op force_cpu params  bug

As the title

As the title

topk chinese doc code example fails as same input names

Add the override function in the elementwise ops

As the title

Co-authored-by: NXing Wu <wuxing03@baidu.com>

Co-authored-by: NXing Wu <wuxing03@baidu.com>

* fix cn doc of warpctc, test=develop, test=document_fix

* fix cn doc of warpctc, test=develop, test=document_fix

Fix the abs doc bug, change the Tenosr to Tensor

* update bn

* update

* update bn