Merge pull request !30592 from lianliguang/code_docs_ms

Merge pull request !30616 from 俞涵/code_docs_1

Merge pull request !30602 from liutongtong9/code_docs_fix_dataset

Merge pull request !30610 from xumengjuan1/code_docs_xx5

Merge pull request !30566 from changzherui/code_docs_add_note

Merge pull request !30568 from Songyuanwei/api

Merge pull request !30469 from wangjun/0223_pp

Merge pull request !30575 from xumengjuan1/code_docs_xx3

Merge pull request !30567 from lichen/code_docs_TransformerRecomputeConfig_API

Merge pull request !30491 from jiahongQian/master

Merge pull request !30494 from YuJianfeng/recompute

Merge pull request !30531 from zhuyuxiao/I4S85V_doc

Merge pull request !29997 from Xiaoda/124-moe-changes

Merge pull request !30512 from zhangyi/code_docs_master

Merge pull request !30432 from changzherui/add_ckpt_check

Merge pull request !30460 from xiaotianci/modify_api_docs

1) extending _Linear's input as 4-dimension tensor: [outer_batch, expert_dim, -1, hidden], and _Liner's BatchMatMul becomes BatchMatMul(4_dim_tensor, 3_dim_tensor);
2) configuring the _Linear's BatchMatMul sharding strategy as [[dp, ep, 1, 1], [ep, 1, mp]];
3) introducing a new parameter 'expert_parallel' in TransformerOpParallelConfig, creating a new class MoEParallelConfig to include 'data_parallel', 'model_parallel' and 'expert_parallel';
4) changing parallel config for FeedForward, TransformerEncoderLayer, TransformerDecoderLayer.

Merge pull request !30329 from luoyang/code_docs_chinese

Merge pull request !30170 from shenwei41/code_docs_chinese

Merge pull request !30374 from jinjiali-kali/code_docs_master

Merge pull request !30269 from zhangfanghe/code_docs_zfh_20220219_nn

Merge pull request !30271 from zhangfanghe/code_docs_zfh_20220219_ops1

Merge pull request !30272 from zhangfanghe/code_docs_zfh_20220219_ops2

Merge pull request !30346 from lianliguang/code_docs_ms