support Glm4Moe

[WYB27]

Description

接入Glm4.5模型
已完成：

• 模型组网已跑通
• 精度已对齐（1e-2级别）
• tp已接入(全部linear)
• sft lora 多卡训练可进行
• pp已接入

todo：
补充单测

[paddle-bot]

Thanks for your contribution!

[lugimzzz]

要仔细验证一下TP & TP+SP收敛精度是否一致

[lugimzzz]

1.验证打开TP+SP是否有问题
2.验证sdpa & flashmask是否有问题
3.验证一下模型保存和热启是否有问题

[codecov-commenter]

Codecov Report

❌ Patch coverage is 18.73684% with 386 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@caee94f). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
paddleformers/transformers/glm4_moe/modeling.py	18.48%	344 Missing ⚠️
...ddleformers/transformers/glm4_moe/configuration.py	15.38%	33 Missing ⚠️
paddleformers/nn/pp_model.py	30.00%	7 Missing ⚠️
paddleformers/trl/llm_utils.py	0.00%	2 Missing ⚠️

❌ Your patch status has failed because the patch coverage (18.73%) is below the target coverage (80.00%). You can increase the patch coverage or adjust the target coverage.

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #2554   +/-   ##
==========================================
  Coverage           ?   28.68%           
==========================================
  Files              ?      308           
  Lines              ?    53764           
  Branches           ?        0           
==========================================
  Hits               ?    15420           
  Misses             ?    38344           
  Partials           ?        0           

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[nepeplwu]

LGTM