* support fuse conv and bn in QAT (#42255)

* support skip_op_list in PostTrainingQuantization (#42378)

* fix unittest

* Fix cond_block_grad error when handle no need grad vras

* Add comment and UT

cherry-pick PR #42777

the doc was editted based on the discussion in the issue:
INT32 Failed on paddle.metric.accuracy: https://github.com/PaddlePaddle/Paddle/issues/42845

* [Dy2Stat]Modify all jit.save path into tempfile

* [Dy2Stat]Modify all jit.save path into tempfile

* put_record_event_in_python_on_timeline_python

* fix

修复paddle.lerp中示例代码错误。

* [MLU] add mlu new profiler (#41138)

* [MLU] add mlu new profiler

* fix format

* [MLU] support add callback to stream (#41831)

* [MLU] add gather mlu kernel (#41969)

* [MLU] add mlu activation kernels (#41751)

    add class NameScopeHelper for adding namescope info
    添加更多 种类优化器状态的映射
    为 IpuStrategy 添加 compilation_progress_logger option 用于输出 编译进度
    部分代码清理和杂项优化

* Reduce time variation for cuda_managed_memory_test (#42458)

* Disable standalone executor for test_tensordot (#42476)

* fix wrong place

* skip bf16 test if not supported (#42503)

* fix PTQ unittest timeout

* fix ut

Co-authored-by: Njoanna.wozna.intel <joanna.wozna@intel.com>

* fix PIL sample mode deprecated warning

* compatible with old pil version

* Fix Release 2.3 Bug

* Fix format

* [Dy2Stat]Fix losting pre/post hook from outermost layer while jit.save

* fix kwargs

* fix unittest

* Extend python einsum interface to make einsum_v2 support multi-operands and switch it to default.

* add opt_einsum dependence

* add yaml and support eager model

* fix by code review

* fix pad3d infer shape

* fix pad3d

* fix pad default value

* fix order

* add unit test

* fix unittest for ci coverage

* add ndhwc check

[cherry-pick 2.3] Add fused_multi_transformer op to optimize transformer generation performance (#42311)

* Add fused_multi_transformer op to optimize transformer generation performance (#41814)

* fix fused_multi_transformer compile failed in cuda arch < sm53 (#42315)

* fix ci timeout

* fix bug of building InferMetaContext (#42211)

* add unitest

* full api fix

* when out is None, go old dygraph mode

* by static check

* first version: support 2-inputs forwards. TODO: 1. backward  2. BroadCast  3. MultiVariable

* time out -> 120

* implement autotune python API

* fix doc

* fix windows error

* fix doc and enable auto-tuning when config is None

* fix windows error

* fix doc

* Optimize the performanece of sum api (#42231)

* optimize the performanece of sum api

* optimize IsDenseTensorInput

* remove debug log

* Add move construct for KernelSignature (#42253)

* add move construct for KernelSignature

* add noexcept

* fix cherry-pick problem

[Eager] Remove retain_grad_flag in accumulation_nade, add is_new_grad args in operator (#42240) (#42290)

fix autotune in reader.py
bug原因：创建dataloader的使用直接使用原始dataloader 等于的方式: new_dataloader = old_dataloader，属于深拷贝，导致后续修改new_dataloader的dataset用于num_workers选择时，会造成old_dataloader的dataset同步被修改。导致模型运行的dataset与实际dataset不相等

ref clipgradbynorm

update

Update __init__.py

cherry-pick from develop PR https://github.com/PaddlePaddle/Paddle/pull/42254