Support PyLayer version of Recompute

[sijunhe]

PR types

PR changes

Description

This pr: PaddlePaddle/Paddle#56793 fix llama generation bug under pylayer version recompute.

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Merging #6885 (bdad639) into develop (a5626f0) will increase coverage by 0.00%.
Report is 1 commits behind head on develop.
The diff coverage is 100.00%.

@@           Coverage Diff            @@
##           develop    #6885   +/-   ##
========================================
  Coverage    59.94%   59.94%           
========================================
  Files          547      547           
  Lines        80996    81001    +5     
========================================
+ Hits         48550    48553    +3     
- Misses       32446    32448    +2     

Files Changed	Coverage Δ
paddlenlp/transformers/chatglm/modeling.py	72.96% <ø> (ø)
paddlenlp/transformers/gpt/modeling.py	66.30% <ø> (ø)
paddlenlp/transformers/bloom/modeling.py	56.34% <100.00%> (+0.05%)	⬆️
paddlenlp/transformers/chatglm_v2/modeling.py	86.82% <100.00%> (+0.03%)	⬆️
paddlenlp/transformers/configuration_utils.py	72.48% <100.00%> (+0.06%)	⬆️
paddlenlp/transformers/llama/modeling.py	68.51% <100.00%> (+0.04%)	⬆️
paddlenlp/transformers/qwen/modeling.py	86.05% <100.00%> (+0.03%)	⬆️

... and 1 file with indirect coverage changes