* change einsum_v2 as default and add new flags: FLAG_einsum_opt=1|0

* make EInsumOP support bf16

* add unittest for BF16

* add condition for test_BF16

* fix bugs

* fix

Co-authored-by: NRyan Jeng <rjeng@nvidia.com>

* support is empty

* fix error

* fix code error

* change to fake empty

* using fake empty first

* using fake empty first

* Support backward prune in fluid

* replace dropout_is_test with is_test.
* improve atol on a100.

* [Eager] fix collective_global_gather

* fix eager_ode = 1

* add embedding yaml

* fix infermeta bug

* fix bug of selected_rows infer_meta

* fix selected_rows

* add unittest

OneDNN md-in-tensor refactoring part 5: Memory descriptor enabled for elementwises, reductions and expand_v2 ops (#43036)

* enabled md in elementwises, reductions and expand_v2

* CI fix for invalid numpy copy

* fixed formatting

* CI rerun

* changes after review

* add fused_bias_dropout_residual_ln op and layer.

* update lite compile cmake

* Update delete_fill_constant_op_pass.cc

* Update analysis_config.cc

* support backward inplace in eager fluid mode

* fix

* fix

* optimize format

* little change

* use original id in dist_op_context.grad_op_id_to_op_id

* del assert

* remove redundant map

* nanmedian op

* 修改cuda kernel的bug

* 修复count_if在其他硬件平台不兼容

* 修复某些cpu硬件不兼容

* 修复某些cpu硬件不兼容

* 修复isnan判断

* 兼容numpy低版本不支持全部nan的情况

* 兼容numpy低版本不支持全部nan的情况

* fix code example

* fix api comment error

* 修改反向传播逻辑以及c++处理逻辑

* 完成修改建议

* typo pre_dim

* update en docs, test=document_fix

* remove numpy in en doc, test=document_fix

* add r,test=document_fix

* 添加api到all

* follow advice from chenwhql

* enhance check for current block docstring for prim2orig interface

* refine if else syntax

* 1st commit

* fix usless change in header transpose_kernel_h file

* add sync

* Fix cond_block_grad error when handle no need grad vras

* Add comment and UT

* addmm supports 1-d input

* fix coverage

* fix

* more ut

* make data transform inplace when tensor is on gpupinned in new dygraph

* fix unittest

* fix alltoall

* rename utest

* support is empty

* fix error

* fix code error

* change to fake empty

* using fake empty first

* using fake empty first

* refactor the optional tensor

* remove optiona<MetaTensor> in InferMeta

* fix bug

* fix optional<vector<Tensor>>

* fix bug

* fix rmsprop

* fix amp of eager_gen

* polish code

* fix deleted code

* fix merge conflict

* polish code

* remove is_nullopt_

* fix merge conflict

* fix merge conflict

* [Dy2Stat]Replace paddle.jit.dy2stat with _jst

* [Dy2Stat]Replace paddle.jit.dy2stat with _jst

* refine code style

* refine code style

* refine trunc uinttest

* refine unittest

* refine ut

* refine fp64 grad check

* experimental nvcc-lazy-module-loading

* remove two empty last line from two files