fix qwen lm_head matrix dimmention alignment

[WAI-clear]

PR types

Bug fixes

PR changes

Models

Description

修复了qwen Torch2Paddle 参数自动转换 lm_head权重参数维度不对齐的问题

[CLAassistant]

All committers have signed the CLA.

[gongel]

LGTM

[codecov]

Codecov Report

Merging #7402 (e1f9fbd) into develop (c013ccb) will decrease coverage by 0.07%.
Report is 7 commits behind head on develop.
The diff coverage is 36.11%.

@@             Coverage Diff             @@
##           develop    #7402      +/-   ##
===========================================
- Coverage    58.60%   58.54%   -0.07%     
===========================================
  Files          574      574              
  Lines        84866    85048     +182     
===========================================
+ Hits         49739    49791      +52     
- Misses       35127    35257     +130     

Files	Coverage Δ
paddlenlp/peft/lora/__init__.py	100.00% <ø> (ø)
paddlenlp/transformers/qwen/modeling.py	81.76% <100.00%> (+0.03%)	⬆️
paddlenlp/transformers/utils.py	68.36% <80.00%> (+1.29%)	⬆️
paddlenlp/utils/serialization.py	89.16% <97.43%> (+0.38%)	⬆️
paddlenlp/transformers/model_utils.py	63.77% <73.33%> (+0.09%)	⬆️
paddlenlp/transformers/llama/modeling.py	67.37% <0.00%> (-0.29%)	⬇️
paddlenlp/peft/lora/lora_model.py	62.50% <15.00%> (-3.45%)	⬇️
paddlenlp/peft/prefix/prefix_model.py	56.52% <15.00%> (-3.57%)	⬇️
paddlenlp/peft/lora/lora_quant_layers.py	38.12% <15.45%> (-49.88%)	⬇️

[paddle-bot]

Thanks for your contribution!