* reduce sum op default fp32, add into amp black list

* reduce_sum default fp32 can avoid return inf when the sum value large than 65504

* change Add to AddN in sum npu op

* add AddInputNames

* change fp16 to fp32 because numpy has accuracy loss in fp16 adding

* delete check

* fix runner error

* refine ptq according comments
* reuse the module to calculate kl threshold

* add amp support for pp

* add amp untest

* fix hdfs and fleet_util for supporting save/load infer;test=develop

* Save all scales to target ops
* Move quant layers to paddle.nn.quant

* delete useless GELU in gelu npu op

* add description

* fix format

* add check_grad in gelu unittest

* refine the old code

* support moving_average_abs_max and per_channel_abs_max

* Add moving_average_abs_max_scale op

* Convert the test program

* add random and prevent deadlock

* dygraph sharding

* update unitest hybrid_parallel_communicate_group

* add get and set for Grad_scaler

* refine some API name and comments

* refine API name and comments

* refine some comments

* added matmul_v2 bf16/fp32 FWD kernel

added matmul_v2 bf16/fp32 FWD kernel

* added formatting

* removed some tests due to timeout in CI

* refactored tests

* merged tests classes into one file

* minor change

* removed test guard for CUDA

* remove skipIf

* changes after review

* formated one file

* minor change

* added skipping UT in CUDA place

* refine temp_scope_vec logic

* polish partial_program

* fix fake var

* add stop_gradient in spec

* fix fake_var

* fix unittest

* support set_device for NPU.

* minor update doc and add more unit test.

* add returns for beamsearchdecoder doc, test=document_fix

* remove heterbox

* remove heterbox