* Add Chinese Doc for cond API

* add doc for gelu

* fix indent, test=document_preview

* fix hyper link, test=document_preview

* fix dtype annotations, test=document_preview

* update gradient comment in Variable API, test=develop

* add sample code for Variable's gradient api, test=develop

test=develop

test=develop

* add Chinese document of control flow API case and switch_case. test=develop

* polish TypeError according to reviews. test=develop

As the title

test=develop

* Add descending parameter for argsort doc.

* Refine doc

* Refine format

* add three new parameters for data_norm
* slot_dim: PaddlePaddle/Paddle#21152
* summary_decay_rate & sync_stats: PaddlePaddle/Paddle#21325

* update adam_op, scale_op, batch_norm_op doc. test=develop

* fix batch_norm example indent. test=develop

* modify the windows inference doc

* Remove the complete code sample and add the corresponding link

* change linux style path to indows style path

* add masked_select api chinese document

* refine, test=document_preview

test=develop

update 1.6.0 inference_lib of Windows to 1.6.1

This reverts commit 6f6cc947.

* modify doc for Attr(data_format) or Attr(data_layout) and add Raises information, test=develop

* modify doc of conv3d_transpose for Raises information, test=develop

* add recompute document

* fix dead links for models

* fix_dead_links

update x2paddle link

* update DetectionMAP reference

* Update DetectionMAP_cn.rst

There are 2 typos at `python_infer_cn.md`.
1. The correct parameter is Float32 instead of kFloat32
2. The correct argument is precision_mode instead of precision

* fix-apis-develop

* fix-one-error

* remove Fake and fix layers_cn.rst