[ZeroPadding] padding to max_length for sequence parallel

[DrownFish19]

PR types

Function optimization

PR changes

Others

Description

This PR addresses the issue where zero padding packs input_ids into a new sequence with varying lengths, which may not be divisible by tensor_parallel_degree. It enables padding the sequence to max_length with a fixed length.

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Attention: Patch coverage is 13.63636% with 19 lines in your changes missing coverage. Please review.

Project coverage is 54.86%. Comparing base (0136312) to head (a41d87b).
Report is 3 commits behind head on develop.

❗ Current head a41d87b differs from pull request most recent head a4b0264

Please upload reports for the commit a4b0264 to get more accurate results.

Files	Patch %	Lines
paddlenlp/datasets/zero_padding_dataset.py	13.63%	19 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #8973      +/-   ##
===========================================
+ Coverage    54.80%   54.86%   +0.05%     
===========================================
  Files          647      647              
  Lines       102462   102475      +13     
===========================================
+ Hits         56157    56223      +66     
+ Misses       46305    46252      -53     

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

[DesmonDay]

LGTM