* fix syncbn

* fix compare op in for in the cuda device

* fix the paddle compare op for the broadcast

* fix spawn default nprocs get error

* polish error message

    paddle.save paddle.static.save protocol的默认值改为4（原默认值为2）。

    pickle protocol=4相交于protocol=2：
        protocol=4时保存/加载大于4G的单个numpy.ndarray 等。
        protocol=4时保存/加载的速度有明显提升。
        Python2 不支持protocol=4（paddle2.1主要支持Python3，不再考虑Python2）。

    兼容问题：pickle版本（protocol）会写到文件里面，pickle load的时候会自动识别到protocol，paddle2.1（paddle.save pickle默认版本为2）可以加载paddle2.1.1的模型（paddle.save pickle默认版本为4）。

原始PR：#32904

Cherry-pick (#33213)

 [cherry-pick][CustomOP]Set GLIBCXX_USE_CXX11_ABI=1 to fix potential GCC ABI problem  (#33153) (#33185)

* Add GLIBCXX_USE_CXX11_ABI flag

* fix typo

* fix typo

* [HybridParallel]Fix pipeline in dygraph (#33007)

* fix pipeline

* fix mp pp dp

* fix utest of hybrid parallel

* add utest for tuple

* fix utest (#33108)

* Support convert sublayers in Sequential Container

* remove paddle.jit.set_code_level

* fix precision of mp

* fix bug of seed

* fix dp

* print group

Change Paddle CI-Cverage Python3.8 Cherry-pick 32515

* [Dy2Static]Add param_guard in ParameterList to support @to_static

* [Dy2Static] Refactor param_guard logic of @to_static (#32867)

* Add param_guard in ParameterList to support @to_static

* Refactor param_guard of @to_static

* fix unittest failed

* add more unittest

* BugFix with ParseInputDataType from LodTensorArray

* BugFix with ParseInputDataType from LodTensorArray

* [Custom Op]Remove PADDLE_WITH_MKLDNN in custom_op  (#32903)

* [Dy2Stat]BugFix StaticAanlysis with gast.Subscript (#32969)

* BugFix StaticAanlysis with gast.Subscript

* remove codes

    test_paddle_save_load 单测随机挂：使用np.ndarray生成随机数组，可能生成nan，造成做对比时结果不匹配（nan != nan）。改为np.random.randn生成随机数组。

    test_paddle_save_load_binary随机挂: 如果一个字符串不能解析为Program，windows上会有超时风险。解决方法：不在windows平台不加载'不能解析为Program的字符串'。

原始PR:#32949

Change Paddle CI-Cverage Python3.8 Cherry-pick 32515

cherry-pick of #32972

* cherrypick for #32640 :add profile and fix dataset hang in heterps;test=develop

* cherrypick for #32640 :add profile and fix dataset hang in heterps;test=develop

* cherrypick for #32640 :add profile and fix dataset hang in heterps;test=develop

* doc of paddle.save/load

* polish doc of paddle.save/load

 Fix CI Python3 on release/2.1 #32930 

* fix dataloader exit hang when join re-enter. test=develop

* double check _shutdown. test=develop

fix error log for reducer

fix doc

fix bug of utest

fix spawn

fix converage

* implement MHA order same as training

* fix fp16 compile issue on old architecture
Co-authored-by: Nzlsh80826 <rewang@nvidia.com>

* remove packages in __all__

* create new public api level paddle.callbacks;paddle.hub;paddle.utils.unique_name

* fix compile error on jetson platform

* remove unused head file

* rm decode_jpeg op on jetson platform

 [CHERRY-PICK2.1]Remove paddle_custom_op dynamic libraries, and link to FLUID_CORE on windows (#32583) (#32769)

* Remove paddle_custom_op dynamic libraries, change link to FLUID_CORE on windows, and check copy_to

* fix CI

* Add casting initializers for bf16 training

* Changes after review

* Correct test and add comment
Co-authored-by: Njoanna.wozna.intel <joanna.wozna@intel.com>

* Add deprecated warning info.

* Add unittest for deprecated decorator.

* Add warning info for tensor.grad

修复了py_layer_op由于没有析构PyLayerContext造成内存(显存)泄露的问题。

原始pr：#32707