Sft flash mask

[wtmlon]

PR types

PR changes

Description

适配 flash mask 数据流，支持 SFT，DPO

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Attention: Patch coverage is 2.94118% with 33 lines in your changes missing coverage. Please review.

Project coverage is 55.61%. Comparing base (3ebe938) to head (1f6bfd1).
Report is 228 commits behind head on develop.

Files with missing lines	Patch %	Lines
paddlenlp/data/data_collator.py	0.00%	11 Missing ⚠️
paddlenlp/trl/trl_data.py	0.00%	10 Missing ⚠️
paddlenlp/datasets/zero_padding_dataset.py	0.00%	4 Missing ⚠️
paddlenlp/trl/dpo_trainer.py	0.00%	4 Missing ⚠️
paddlenlp/transformers/llama/fusion_ops.py	0.00%	2 Missing ⚠️
paddlenlp/transformers/llama/modeling.py	33.33%	2 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #8664      +/-   ##
===========================================
- Coverage    55.62%   55.61%   -0.01%     
===========================================
  Files          620      620              
  Lines        96949    96965      +16     
===========================================
+ Hits         53929    53930       +1     
- Misses       43020    43035      +15     

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

[lugimzzz]

lgtm