[Bug fixes] fix TP_of_chatglm

[SevenSamon]

PR types

Bug fixes

PR changes

Models

Description

• The MLP.dense_h_to_4h segmentation of chatglmV1 is based on the default segmentation, resulting in the activation operation and the single card not aligned.
• Customize the segmentation strategy function in _get_tensor_parallel_mappings

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Attention: Patch coverage is 7.14286% with 26 lines in your changes missing coverage. Please review.

Project coverage is 53.25%. Comparing base (e0d2809) to head (3b8f2ab).
Report is 114 commits behind head on develop.

Files with missing lines	Patch %	Lines
paddlenlp/transformers/chatglm/modeling.py	7.14%	26 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #8949      +/-   ##
===========================================
- Coverage    54.79%   53.25%   -1.55%     
===========================================
  Files          636      652      +16     
  Lines        99876   105642    +5766     
===========================================
+ Hits         54732    56255    +1523     
- Misses       45144    49387    +4243     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ZHUI]

PR还需要吗?

[CLAassistant]

All committers have signed the CLA.