* Revert "Align CTC grad scale same with ESPNet (#34729)"

This reverts commit 10f9644c.

* ctc grad compute on gpu

* Fix bug when pre_layer_norm is false.

* Add WIP version of elementwise_div_mkldnn without working dy grad

* Add dy gradient calculation implementation, disable broadcast tests

* Readd removed tests from static_mode_white_list

* Add bfloat16 gradient tests, remove int8 and uint8 support

* - Change the way dy grad is calculated to improve performance
- Refactor BinaryMKLDNNHandler to use a default parameter

* Change copyright year

* Refactor as suggested

* Attempt to bypass CI Approval
not accepting max_relative_error

* Fix formatting issue

* WIP: add cache

* delete move constructor and operator= for CuFFTHandle and FFTConfig

* remove log from CuFFTHandle and FFTConfig

* add lrucache for fft rocm backend

* disable LRUCache when CUFFT_VERSION >= 10200

* disbale copy and move for hipFFTHandle; format code

* clean debug code
Co-authored-by: NXiaoxu Chen <chenxx_id@163.com>

* fix matmul dim error

* fix wrong dim check in matmul

* fix fftshift/ifftshift on static mode
* update roll_op version
* add more test cases for fftshift/ifftshift

* add eigvalsh with is_test

* add eigvalsh op

* fix backward bug

* forward and backward, float and complex, unittest

* remove eigvalsh_helper.h

* remove changes of cusolver.h

* fix unittest

* fix unittest bug

* update code following eigh

* fix test

* update lapack

* pull develop

* update funcor

* fix unittest bug

* fix details

* add tensor_method_func

* fix notes

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；

将 #35905 和 #35843 PR中新增的的python api接口移到incubate目录下。

* [NPU] fix argsort op, test=develop

* remove debug files, test=develop

* fix typo, test=develop

* address review comments, test=develop

* optimize fast tokenizer

* [NPU] modifications for model ernie-1.0

* rollback 503003 and change cast to dtype

这个PR是fused_feedforward反向的代码

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。

* Add bincount op

* upload cpu version

* fix unitest

* fix unittest

* fix unittest

* fix en doc

* add more test

* fix en doc

* add more test case

* fix test

* fix input vailidation

* fix input check

* fix unittest

* fix test

* fix en doc

* add some ops to train ssd on kunlun

* add some ops to train ssd on kunlun

* add some ops to train ssd on kunlun

* update cast op unittest

* update cast op unittest

* update cast op unittest

* update xpu cmake

* update cast unittest

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；

* Support elementwise_add triple grad Kernel

* Change code-format to follow CI std

* Removed unreasonable code, and fixed an input uninitialized issue

* Support elementwise_add triple grad Kernel

* Change code-format to follow CI std

* Removed unreasonable code, and fixed an input uninitialized issue

* added base changes for matmul_v2+trans+resh fuse pass

* added full matmul_v2+transpose+reshape pass

* removed a file added by mistake

* added reviewers suggestions

* Changed ops type in checking capatibility version

* Deteled one statement

* add sync_batch_norm (support train, infer, and fp32, fp16, and NCHW, NHWC)

* [NPU] Delete debug codes

* [NPU] Remove FP16

* add viterbi decode cpu kernel

* add viterbi decoder api in paddle.text

* add a data buffer once to avoid create many small pieces of data buffer frequently

* fix viterbi max_seq_length bug

* fix seq_len=1 bug

* fix device context

* move split out of for loop

* remove INVERSE_SUB

* remove 2 GET_CAST_MASK

* remove 1 loop

* remove Functor

* add to_static deploy code

* use MAX_FUNC instead of ELE_MAX

* add MaxFunctor

* impl max_func

* remove MaxFunctor

* remove cast op

* use REGISTER_OP_WITHOUT_GRADIENT

* add viterbi cuda kernel

* add FIX_BLOCKDIM_CASE macro

* add MKL add, mul; add get data mask

* add arange mkl impl

* add CPU Argmax

* add cpu gather

* use EXECUTE_MKL_ELEMENT_BINARY_OP instead of some ADD, MUL

* use SameDimsBinaryOP instead of EXECUTE_MKL_ELEMENT_BINARY_OP

* use SAME_DIMS_ELEMENT_BINARY_OP

* add SimpleBroadcastBinaryOP

* use int instead of int64_t to accelerate

* optimize SimpleBroadcastBinaryOP

* optimize SimpleBroadcastBinaryOP

* optimize performance in both single thread and multithread situation

* remove useless line

* remove useless code

* add CREATE_TENSOR_BUFFER macro

* add INIT_REQUIRED_TENSOR macro

* add comment

* fix windows ci

* add viterbi unittest

* remove cuda add functor

* remove cuda equal

* remove a template function

* fix windows ci

* fix windows dtype

* remove some template instance

* remove useless header file

* remove some blockdim

* remove transpose impl

* accelerate cpu performance on single thread situation

* viterbi_decode->crf_decode

* rename crf params name

* add viterbi api test

* remove useless import

* add enable_static

* use viterbi decoder

* fix viterbi len=1

* fix  viterbi unittest

* remove useless comments

* reconstruct viterbi decode

* remove ADD,SUB,MUL structure

* fix coverage

* remove CREATE_TENSOR

* add name args

* crf.py->ops.py; with_start_stop_tag->include_start_end_tag

* update crf_decode en docs

* fix viterbi decode en docs

* fix some review comments

* add FIXED_BLOCK_DIM_CASE in cuda

* push_back->emplace_back

* crf_decode->viterbi_decode; include_start_end_tag->include_bos_eos_tag

* paddle.text.ops.viterbi_decode->paddle.text.viterbi_decode

* fix viterbi_decode en docs

* add some ops to train ssd on kunlun

* update test_fill_any_like_op_xpu.py

* Update the implement of reduceAnyKernel according to kernel primitive api
* Fix a bug in ReadData, ReadDataBc and ReadDataReduce when NX != 1

* fix global gather and global scatter operators

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.