这个PR只包含fused_feedforward前向的代码。

相关kernel实现：fused_dropout_act_bias, fused_residual_dropout_bias, fused_layernorm_residual_dropout_bias

fused_feedforward是一个融合算子，该算子对transformer模型的feed forward层的算子进行融合和封装，使得前端只呈现一个接口，通过融合减少部分访存和kernel launch的时间，以此提升性能。

功能：本PR的目标是提高attention模块的计算性能。
为了减少框架层对op的调度开销，本PR通过在C++层手动实现attention模块，对外提供attention 大op；
为了减少防存开销，本PR采取了两种优化方法：
（1）在q,k,v计算时通过共享输入X，将该处的gemm，transpose和bias add从三次调用减少为一次；
（2）使用kernel融合优化技术，在不同cuda kernel之间通过寄存器传输数据；

Add Tokenizer related functionalities for Transformer model in order that the process of training and predicting is consistent.

* support the text string as an input Tensor
* support the "VOCAB"unordered_map<wstring, int> as an input Tensor to lookup tokens
* Tokenizer used for BERT. This tokenizer applies an end-to-end, text string to wordpiece tokenization.
* It first applies basic tokenization, followed by wordpiece tokenization.

* add func/class API psroi_pool and UT

* add UT in static mode

* Remove redundant type checks in static mode

* More detailed description for test_psroi_pool_op

* fix code format of UT

* fix en-doc

* adam to adamw in AdamW

* add lr_ratio in adamw

* refine logic bug in cpu adamw

* delete fix bug for cpu adamw

* delete fix bug for cpu adamw

* add pure fp16 major function in auto_cast & tracer

* support master weight in dygraph for pure fp16

* check mix dtype of fp16&fp32 for check_finite_and_unscale op

* change pure fp16 funtion name

* refine some bug in auto_cast

* refine auto_cast interface logic

* add param _casted_by_pure_fp16 for class Layer

* support state_dict hook for save model by user appointed dtype in pure_fp16_decorator

* refine pure_fp16_decorator as decorator

* add unittest

* add comment

* add comment

* support recompute

* add comment for auto_cast and decorator

* support to_static_state_dict for paddle.jit.save

* unlimite models num and optimizers num

* add lookup_table in black_list

* fix momentum and layer state_dict

* fix bug in layer state_dict

* fix bug in layer state_dict_helper

* refine unittest

* refine test_momentun_op

* refine interface and some code

* refine amp_decorator interface

* refine pure fp16 interface

* refine master weight interface

* init matrix_rank op, add matrix_rank CPU code and test

* add GPU kernel, remove svd_eigen.h

* add CPU kernel when tol is tensor

* add cpu and gpu code when tol is tensor

* fix CI-ROCM error

* add matrix_rank API describe, fix PR-CI-Py3 error

* fix PR-CI-Windows error, add matrix_rank API test

* delete useless comments

* fix review

* add my code in svd_helper.h

* update doc commets

* remove spaces

* sparse_momentum_op is used to save w@GRAD memory for gather_op when gather from a large parameter

* adamw support cuda

* adamw support cuda

* add unique_consecutive_op

* add unique_consecutive_op

* add unique_consecutive_op

* add unique_consecutive_op

* add unique_consecutive_op

* add unique_consecutive_op

* add unique_consecutive_op

* add unique_consecutive_op

* remove unity build

* add unique_consecutive op

* add unique_consecutive op

* add enable static

* add noqa

* add space line

* add default case.

* add comma

* add space line

* modify unique_consecutive unittest

* optimize ut coverage

* rebase develop

* improve coverage

* update en docs

* update en docs

* update en docs

* update en docs

* update en docs

* update en doc

* add recompute for pp

* add recompute offload

* add recompute partition

* add scatter-gather opt

* fix topo for pp

* rename function

* fix bug about deallocating None, test=develop

* use PYTHON_C_API in dygraph, test=develop

* Modify into core.ops.run_program

* add DDout in core.ops.run_program

* fix typo

* add DOut

* fix typo

* put DOut last

* add sendrecv, test=develop

* fix model parallel

* rm parallel_help.py

* add embedding

* add allreduce and broadcast without test (#31024)

add allreduce and broadcast without test

* Refactor HCCLCommContext to be compatible with Paddle (#31359)

Refactor HCCLCommContext to be compatible with Paddle (#31359)

* [NPU] add npu kernel for communication op (#31437)

* add allreduce and broadcast without test

* add c_broadcast_test case

* build c_comm_init and c_create_group operators

* make the whole thing compile

* add broadcast and init op test case but run failed

* make unit test compile

* fix broadcast test bug and change into hcom for ccl

* change c_comm_init and c_create_group ops accordingly

* make tests compile

* transfer code to 27

* compiled successfully in 28, but run failed

* test broadcast in 28, but failed

* make hcom primitives work

* change hccl data type for base.h

* fix broadcast bug

* make attributes work

* fix group name bug

* add allreduce but test failed

* allreduce bug for qiuliang

* allreduce finished

* add allgather and reducescatter

* merge all op code

* add allgather test

* finish run all ccl op test exclude send/recv

* all all op and test exclude send/recv

* send_v2_npu.cc recv_v2_npiu.cc compiled

* fix ccl core dump bug and test allgather, reducescatter, broadcast op

* fix allreduce bug just for test

* hcom send&recv test pass, without hcom_destroy

* for qiuliang test

* Ascend Send&Recv Test Pass

* all op (ex send/recv) ok

* fix bug

* merge all ccl op

* style merge to PaddlePaddle

* merge style

* new merge style

* merge style 2

* insert an empty at the end

* disable ctest for hcom to pass ci
Co-authored-by: Nvoid-main <voidmain1313113@gmail.com>
Co-authored-by: Nf2hkop <f2huestc@outlook.com>

* Add auto-increasing tag id for Hcom OPs (#31702)

* add c_reduce_sum op (#31793)

add c_reduce_sum op

* update Ascendrc hccl to 20.3 (#32126)

update Ascendrc hccl to 20.3 (#32126)

* fix merge code

* change cmake.txt1

* [NPU] Support npu kernel for c sync stream op (#31386)

* sync stream npu op

* add with_ascend_acl

* update c++ unittest

* compile all failed

* try to pre commit

* after pre commit

* merge&compile&test hccl successfully!

* fix code style

* fix code style

* fix bugs about hccl

* fix some bugs

* fix code style

* fix style

* fix style

* fix

* fixed

* merge develop
Co-authored-by: Nlw921014 <liuwei921014@yeah.net>
Co-authored-by: NVoid Main <voidmain1313113@gmail.com>
Co-authored-by: Nf2hkop <f2huestc@outlook.com>
Co-authored-by: Nxiayanming <41795079@qq.com>

* Ascend rc (#30483)

* Fix compilcation on CANN20.1 and older (#30494)

Fix compilcation on CANN20.1 and older

* Add distribution supported (#30578)

Add distribution supported

* Build praser for Hcom* operators (#30627)

Build praser for Hcom* operators

* Pass device_ids info from launch to trainer. (#30632)

Pass device_ids info from launch to trainer

* Add Hccl program group (#30642)

Add Hccl program group

* Add startup bash files of test_ascend_group. (#30645)

Add startup bash files of test_ascend_group

* cleanup (#30646)

cleanup test_ascend_group.py

* [Feature] Build parser to support distributed training (#30658)

[Feature] Build parser to support distributed training

* fix compilation on ascend-20.1 (#30722)

fix compilation on ascend-20.1

* Dev/fix ascend string (#30749)

Dev/fix ascend string

* code style (#30781)

code style

* Merge ascend_optimizer and ascend_parser. (#30776)

Merge ascend_optimizer and ascend_parser.

* Ascendrc add converted op : [range/equal/range/uniform_random/expand/squeeze], fix cast op bug  (#30797)

Ascendrc add converted op : [range/equal/range/uniform_random/expand/squeeze], fix cast op bug

* Add paddle ascend distribution training supported (#30796)

Add paddle ascend distribution training supported

* pass cxx_flags to gloo cmake (#30857)

* Destroy session first. (#30954)

Destroy session first.

* merge

* fix, test=develop

* fix, test=develop

* fix style, test=develop

* fix, test=develop

* fix

* fix log fatal, test=develop

* fix enforce style, test=develop

* fix, test=develop

* fix, test=develop

* fix rccl, test=develop

* fix test, test=develop

* fix, test=develop

* fix, test=develop

* fix, test=develop

* fix node_num, test=develop

* fix ids str, test=develop

* fix ids str, test=develop

* fix ids str, test=develop

* fix, test=develop

* fix, test=develop

* fix, test=develop

* fix, test=develop

* fix, test=develop

* fix, test=develop

* fix, test=develop

* fix, test=develop

* fix style code, test=develop

* fix style code, test=develop

* fix style code, test=develop

* fix style code, test=develop
Co-authored-by: Nhutuxian <hutuxian2011@sina.cn>
Co-authored-by: Ngongweibao <weibao.gong@gmail.com>
Co-authored-by: NVoid Main <voidmain1313113@gmail.com>
Co-authored-by: NLeo Chen <chenqiuliang@baidu.com>
Co-authored-by: Ndingsiyu <18369187719@163.com>
Co-authored-by: NOleNet <olenet@126.com>

* new group

* ci compatible fix

* assert nccl

* Use layer to calculate output scale
* add backward for moving_average_abs_max_scale and save output scales to op's attr

* Modify the calculation logic of LambOptimizer

* add view strategy on squeeze,unsqueeze,reshape,flatten

* add squeeze unittest

* add unittests

* use View strategy as name rather than Reuse Allacation

* fix view api doc

* fix format

* use core.ops when input of reshape2 is Tensor

* fix test_cross_entropy_loss error because of reshape2

* fix test_cross_entropy_loss error because of reshape2

* add inplace strategy

* add elementwise_add sub

* let backward op not use inplace

* grad op do not use inplace

* fix memory increase error and add leaf error message

* delete selected_rows

* change op_function

* little change

* solve HandleViewBetweenInputAndOutput

* add unittest and leaf error message

* merge view error

* optimize op_function_generator format and support sum inplace op

* fix format of basic_engine

* fix format for framework

* little change of variable wrapper

* add reshape, squeeze, unsqueeze, scatter api

* add relu elu tanh softmax inplace api

* fix test_squeeze_op unittest

* fix test_relu_op unittest

* fix comment problems

* delete sample code of inplace api

* add reference of grad_pending_nodes in basic_engine

* fix unittest name

* add inplace apis into wlist

* fix error message

* add PADDLE_ENFORCE for set grad op twice

* fix head file error

* add view strategy on squeeze,unsqueeze,reshape,flatten

* add squeeze unittest

* add unittests

* use View strategy as name rather than Reuse Allacation

* fix view api doc

* fix format

* use core.ops when input of reshape2 is Tensor

* fix test_cross_entropy_loss error because of reshape2

* delete selected_rows

* change op_function

* little change

* solve HandleViewBetweenInputAndOutput

* fix dtype of ungenerated grad var

* update ut

* refine code

* set default dtype

* fix could_use_cudnn bug

* remove debug code

* re-implement

* fix bug

* test=develop, add model_average and lookahead

* add dispenable input 'shape' for core.ops.reshape2

* add dispenable inputs for core.ops.reshape2/expand/slice

* add ut

* add the weight decay func for the momentum op

* Add the multi_precision function in Momentum Optimizer.

* Make sure that the initial value of master weights are same with the fp16 weights.

* add static loss scaling.

* add the rescale_grad function in the pure fp16 training.

* use the original momentum updating method.

* Polish some codes, such as variable names.

* add docstring for apis.

* update the var creation details of _create_master_weight.

* not modify codes about imperative momentum updating.

* Fix the error of test_dist_sparse_tensor_load_momentum UT.

* add unit test for multi precision fp16 training.

* add more unit tests for CI.

* Use lower threshold values for allclose comparing in test_multi_precision_fp16_train UT.

* For CI Coverage Checking.

* Add check for non-dispensable input

* fix typo

* add generate_proposals_v2 op

* add op_function_generator for box_coder

* fix format

* Enhance multiclass_nms to support LoD for dygraph mode

* fix some error in multiclass_nms

* update GetLodFromRoisNum to GetNmsLodFromRoisNum

* update matrix nms op to api 2.0

* modify code according to review

* Implement the function of OueScaleForTraining/OutScaleForInference in dygraph

test=develop

* add float64 input to ctc_loss

* modified error message of  warpctc

* update repo and tag of warpctc

* add test for warpctc with float64 input

* modified warpctc.cmake to make sure build always

* resolved sample code bug of warpctc

* add core.ops in warpctc dygraph

* fix a bug of test

* Finished ChannelWiseQuantDequantAbsMaxOp and Passed unittests.

* Finished channel-wise quantize strategy in imperative quantization.

* Added Cuda code of ChannelWiseQuantDequantMaxAbsOP
Add Cuda code of ChannelWiseQuantDequantMaxAbsOp

* Add quant_axis for channel_wise quant.

* fixed a bug in unnitests, which will not trigger axis = 1 case and cannot meet the coverage rate requirement.

* Added some assert infomation and fixed some coding style mistakes.