[DCU] fix llama inference bug on DCU

[Deleter-D]

PR types

Bug fixes

PR changes

Models

Description

Some changes in PR 8800 affect the function of DCU in other inference scenarios, so restrictions are added to reduce the scope of impact.

The following scenarios have been tested on the DCU

• dynamic
  • fp16
  • a8w8
  • a8w8c8
• static
  • fp16
  • a8w8
  • a8w8c8

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Attention: Patch coverage is 0% with 2 lines in your changes missing coverage. Please review.

Project coverage is 55.51%. Comparing base (5fd6dd2) to head (01163e0).
Report is 226 commits behind head on develop.

Files with missing lines	Patch %	Lines
...erimental/transformers/fused_transformer_layers.py	0.00%	1 Missing ⚠️
...dlenlp/experimental/transformers/llama/modeling.py	0.00%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #8815      +/-   ##
===========================================
+ Coverage    55.47%   55.51%   +0.04%     
===========================================
  Files          631      631              
  Lines        98468    99126     +658     
===========================================
+ Hits         54621    55031     +410     
- Misses       43847    44095     +248     

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

[DesmonDay]

LGTM