* support OD level and skip dynamic loss scaling for bf16

* add flags for phi

* fix compile bugs

* fix ci bugs

* fix inference bugs

* fix cinn' bugs

* fix cinn bugs

* perfect code according comment

* fix ci bugs

* fix ci bugs

* print modifed flags

* fix ref, opt print

* fix default getter

* fix ut

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* test=allcase

* fix doc erros, test=allcase

* test,test=develop

* test,test=develop

* test,test=develop

* test,test=develop

* test,test=develop

* test,test=develop

* test,test=develop

* test ,test=develop

* [CustomDevice] add pipeline parallel support

* update

* update

* apply gcc12 to gpups

* apply gcc12 to gpups

* apply gcc12 to gpups

* apply gcc12 to gpups

* apply gcc12 to gpups

* apply gcc12 to gpups

* apply gcc12 to gpips

* apply gcc12 to gpups

* apply gcc12 to gpups

* test

* test

* apply gcc12 to gpups

* apply_gcc12_to_gpups

* fix compiler bug

* fix compiler bug

* test

* fix dangling-pointer compiler

* fix dangling-pointer compiler

* fix dangling-pointer compiler

* apply_gcc12_to_gpups

* apply gcc12 to gpups

* Update cuda_streams_py.cc

* Delete temp param in eager_gen

* support 0-D output and 0-D as indice in __getitem__

* fix tests

* fix inference and UT

* add unittest for setitem

* fix xpu test

* fix xpu 0-d

* [CustomOP error] Add attrs type check

* fix global variable order bug

* include unordered_set

* fix ParseAttrStr compile error

* [CustomDevice] add recompute support

* update

* cherry-pick fleet executor from 2.4

* fix test case

1. modify set_value op, use Scalars to represent attr `values`, instead of a bunch of attributs of various types; (#52408)

2. add program converter and set_value op as an example, which provides the functionality to convert `paddle::framework::ProgramDesc` between old and new formats(the differences are mainly some operators with incompatible updates in the definition);
3. program version and operator version map now are always saved when serializing `paddle::framework::ProgramDesc` to identify the version;
3. provide an option `legacy_format=false` in  serialization of `paddle::framework::ProgramDesc`, it decided whether to convert ProgramDesc back to a legacy format, which is compatible for paddle 2.4.2 or earlier versions to load and execute;
4. deserialization of `paddle::framework::ProgramDesc` is now automatically detecting whether the bytes it receives is in legacy format(contains any of the operators that has been incompatibly updated and have any attribute of type `Scalar`) and convert it to new format. But if you want a faithful deserialization without the automatic conversion, you can use protobuf's deserialization instead. Though it is not recommended, it can be used for the purpose of testing.

* support set master_grad

* move register_hook to auto_cast

* update unittest

* fix fp16 test

* update for review comments

* [Opt Performance] Optimize custom operator performance, reconstruct python API auto-gen, add cache and use const inference

* opt AutoGradMeta implementation

* remove profiler codes

* fix unit test

* change year, 2021->2023

* fix int64_t parse bug

* add strategy force_sequential_run

* remove flag

* fix

* fix

* fix

* fix

* fix

* fix

* fix

* fix

* fix

* mv WITH_ASCEND_CL

* mv WITH_ASCEND

* rollback

* remove WITH_ASCEND

* remove WITH_ASCEND

* add strategy force_sequential_run

* fix

* fix

* fix

* fix

* fix

* mv WITH_ASCEND_CL

* mv WITH_ASCEND

* rollback

* rem is_compiled_with_npu

* rem nup related code

* make lint happy

* rem test

* remove some tests

* Update grad_scaler.py

* fix an error