[AutoParallel] Sequence Parallel for Auto Mode

[JZ-LIANG]

PR types

Function optimization

PR changes

Models

Description

Sequence Parallel for Auto Mode
受限于当前标记补全模块的问题，有3处地方的标记不够直观，未来可以更完善：

1. 用户接口处暂不支持 SP 独立维度和通信组，SP 绑定 TP (和动态图 config 对齐)
2. row linear matmul 的标记： shard_op 不支持 linear 等(module)，此处的标记需要调用paddle framework 内部函数
3. transfermer layer input 的标记： SP 标记需要考虑DP 的标记防止冲突，出于较保守精度对齐考虑的，DP 在跨mesh 时需要重新标记但这影响了SP 信息的 propagation

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Merging #7301 (79e1a3e) into develop (f9021f8) will decrease coverage by 0.27%.
Report is 32 commits behind head on develop.
The diff coverage is n/a.

❗ Current head 79e1a3e differs from pull request most recent head 140b8e2. Consider uploading reports for the commit 140b8e2 to get more accurate results

@@             Coverage Diff             @@
##           develop    #7301      +/-   ##
===========================================
- Coverage    59.21%   58.94%   -0.27%     
===========================================
  Files          568      573       +5     
  Lines        83605    84228     +623     
===========================================
+ Hits         49504    49646     +142     
- Misses       34101    34582     +481     

see 32 files with indirect coverage changes