Fix shared weights sync for PipelineLayer

[DrownFish19]

PR types

Bug fixes

PR changes

Others

Description

1. PipelineLayer will create parameters for each layer and call _synchronize_shared_weights() to synchronize the shared parameters.
2. When setting the model state_dict, _synchronize_shared_weights will be called to synchronize the shared parameters. However, self._init_weights will re-initialize the parameters without synchronizing the shared parameters. If the following step does not load a checkpoint, the shared parameters will be different.

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Attention: 4 lines in your changes are missing coverage. Please review.

Comparison is base (adf9e6f) 57.30% compared to head (2abb57b) 57.30%.
Report is 8 commits behind head on develop.

Files	Patch %	Lines
paddlenlp/transformers/llama/modeling_auto.py	0.00%	2 Missing ⚠️
paddlenlp/trainer/training_args.py	0.00%	1 Missing ⚠️
paddlenlp/transformers/model_utils.py	66.66%	1 Missing ⚠️

Additional details and impacted files

@@ Coverage Diff @@ ## develop #7772 +/- ## =========================================== - Coverage 57.30% 57.30% -0.01%  =========================================== Files 584 584 Lines 87688 87690 +2 =========================================== + Hits 50252 50253 +1  - Misses 37436 37437 +1 

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

[wawltor]

LGTM