Merge pull request !3285 from linqingke/gpu_ops

Merge pull request !3318 from liyong126/fix_save_doc

Merge pull request !3331 from liuxiao93/Add-BNInference-opInfo

Merge pull request !3328 from shenwei41/sw_master

Merge pull request !3289 from ZhangQinghua/master

Merge pull request !3342 from cathwong/ckw_c_api_skip

Merge pull request !3294 from nsyca/removal_pass

Merge pull request !3303 from ZiruiWu/fix_vocab_validator

Merge pull request !3336 from vlne-v1/remove_loop_can_unroll

Merge pull request !3337 from zhangbuxue/add_check_for_implict_conversion_when_scalar_is_string

Merge pull request !3332 from jiangzhiwen/fix_csv_shard

Merge pull request !3309 from lianliguang/master

Merge pull request !3319 from gengdongjie/master

!3335 scripts updation for bert to debug failures of pre-training processing when warmup step was set to be zero
Merge pull request !3335 from shibeiji/bert_script_debug

Merge pull request !3330 from lichen/autoparallel_support_dataset_in_gpu

Merge pull request !3327 from wangnan39/fix_bug_in_sparse_proximal_adam_optim

Merge pull request !3295 from wangqiuliang/support-mix-precision

Merge pull request !3266 from zhaoting/ResNext50

Merge pull request !3306 from guozhijian/fix_api_example

Merge pull request !3322 from dinghao/master

Merge pull request !3321 from anzhengqi/fix-wide-deep-performance

!2410 Support insert memcpy between two hccl op if the part outputs of prior hccl op linking to next hccl op
Merge pull request !2410 from huanghui/insert-memcpy-async-pass

Merge pull request !3304 from jinyaohui/master

Merge pull request !3315 from yao_yf/restore_auto_parallel_reshape_ut

Merge pull request !3273 from changzherui/save_slice_tensor

Merge pull request !3180 from alashkari/update-aug-ops-2