* fusedAttenGrad_noGrad

* code style fix

* add ut

* remove unnecessary log

* rm legacy layers part6

* rm non_static_mode

* modify non_static

* minor change

* rm loss

* rm in_legacy part8

* minor change

* rm legacy nn part2

* rm _non_static_mode

* modify

* modify unpool test

* modify unpool test

* modify loss

* keep legacy for layer_norm

* Add CheckPRTemplate.py;test=document_fix;test=prtemplate

* Add CheckPRTemplate.py;test=document_fix;test=prtemplate

* Add CheckPRTemplate.py;test=document_fix;test=prtemplate

remove fluid.contrib.fused_elemwise_activation, sequence_topk_avg_pooling, var_conv_2d, match_matrix_tensor and tree_conv (#49331)

* add skip run

* alloc minimum memory

* skip check_size in Alloc

* skip check_size in Alloc

* skip check_size in Alloc

* fix cases when tensor is initialized or empty

* alloc empty output for place info

* add test

* increase timeout

* format code

* skip cpu

* add cudnn_deterministic

* fit for hostAlloc

* follow comments

* change check_size to fake_alloc

* generate the static op of some ops

* add the VERSION of pixel_shuffle

* change the API doc of isclose

* change the API doc of isclose

* fix the isclose op comment

* einsum support 0d tensor.
1. support 0d tensor in multi-operands.
2. add 9 unittests for einsum 0d tensor.

* override NVIDIA_TF32_OVERRIDE to avoid accuracy problem in 11.2 and 11.8

* update pypi doc

* update pypi doc

* update pypi doc

* empty commit, re-trigger all ci
Co-authored-by: NSigureMo <sigure.qaq@gmail.com>

* [AutoParallel] adapt for clip

* fix unittest

* enable_static

* fix dist_fill_constant_batch_size_like

* fix process_mesh.shape

* update cond of modifying shape_list

* rm legacy fluid part4

* rm non_static_mode

* minor change

* modify initializer

* rm legacy for initializer

* fix dataloader test

This PR increased the delta in unit test for CUDA 11.8. The reason of this fix:
(1) It seems CUDA 11.8 has higher delta in accuracy result. Our other targets for seresnext under parallel executor have already added delta such as CPU, all reduce test cases, so we did same for GPU base case with CUDA 11.8
(2) A new executor is under developing in PaddlePaddle team, so the unit test for old executor can be relaxed.

* delete old dygraph pylayer

* fix CINN should add float16.h may install bug

* reupdate setuppy support float16

* add only if float16.h file exists

* fix run_setup problem

* test

* fix fold for large bs

* fix fold for large bs

This reverts commit e1e8bf72.

* fix input order

* add unittest

* update cmakelist

* [AutoParallel] quantization pass support export

* support subgraph

* move_presist_var_to_global_block

* update unittest

* fix ci-coverage

* fix codestyle

* fix fake_dequantize_op

* remove unused var

* fix ci error and aprroval error

* add unittest for fp16 in test_dequant_linear

* replace mutable data

* fix unittest in non-cuda-core

* fix unittest
Co-authored-by: Ncarryyu <569782149@qq.com>
Co-authored-by: Nwufeisheng <wfs1997@163.com>

* delete old dygraph sharding

* cinn use interpretercore

* fix bug

* fix compile bug

* fix scope bug

* refine code

* refine code by comment

* refine code by comment

* Support priority scheduling for standalone executor

* Add CPU test

* rm in_legacy_dygraph nn part1

* rm non_static_mode

* modify rrelu

* delete _in_legacy_dygraph