【Qwen2-VL Inference】add qwen2-vl high performance inference

[chang-wenbin]

PR types

Performance optimization

PR changes

Models

Description

add qwen2-vl high performance inference

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Attention: Patch coverage is 2.50000% with 39 lines in your changes missing coverage. Please review.

Project coverage is 52.80%. Comparing base (9eb3cfe) to head (f2fbfba).
Report is 6 commits behind head on develop.

Files with missing lines	Patch %	Lines
...dlenlp/experimental/transformers/qwen2/modeling.py	0.00%	38 Missing ⚠️
paddlenlp/transformers/auto/configuration.py	50.00%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #9575      +/-   ##
===========================================
- Coverage    52.81%   52.80%   -0.01%     
===========================================
  Files          710      710              
  Lines       111234   111247      +13     
===========================================
- Hits         58744    58743       -1     
- Misses       52490    52504      +14     

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

[zhoutianzi666]

这里也加上说明，告诉别人为什么这样改