[LLM] Update sequence parallel linear import

[DrownFish19]

PR types

Function optimization

PR changes

Models

Description

1. All models imports ColumnSequenceParallelLinear and RowSequenceParallelLinear from linear_utils, where ColumnSequenceParallelLinear and RowSequenceParallelLinear are defined for different deivces including gpu, npu and xpu.
2. nn.Linear is replace by linear_utils.Linear. When device is gpu or npu, nn.Linear is equal to linear_utils.Linear. When device is 'xpu', linear_utils.Linear is designed for xpu based on nn.Linear, which can achieve better performance.

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Attention: Patch coverage is 66.14173% with 43 lines in your changes missing coverage. Please review.

Project coverage is 55.70%. Comparing base (fbe613b) to head (76be848).
Report is 214 commits behind head on develop.

Files with missing lines	Patch %	Lines
paddlenlp/transformers/gemma/modeling.py	68.42%	6 Missing ⚠️
paddlenlp/transformers/linear_utils.py	0.00%	6 Missing ⚠️
paddlenlp/transformers/qwen2/modeling.py	73.91%	6 Missing ⚠️
paddlenlp/transformers/qwen2_moe/modeling.py	53.84%	6 Missing ⚠️
paddlenlp/peft/lora/lora_layers.py	28.57%	5 Missing ⚠️
paddlenlp/transformers/gpt/modeling.py	76.19%	5 Missing ⚠️
paddlenlp/peft/lora/lora_model.py	55.55%	4 Missing ⚠️
paddlenlp/transformers/mixtral/modeling.py	81.81%	4 Missing ⚠️
paddlenlp/transformers/qwen/modeling.py	85.71%	1 Missing ⚠️

Additional details and impacted files

@@ Coverage Diff @@ ## develop #8706 +/- ## =========================================== - Coverage 55.70% 55.70% -0.01%  =========================================== Files 623 623 Lines 97399 97424 +25 =========================================== + Hits 54255 54268 +13  - Misses 43144 43156 +12 

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