[XPU] add xpu support for llama sft

[tizhou86]

PR types

New features

PR changes

Models

Description

• Llama model supports XPU
• 简要说明：XPU自定义的融合算子通过Paddle C++ Extension的方式接入，XPU自定义Paddle算子库的名称叫做paddle_xpu(aka. fast_paddle)
• 相关PR参考：[XPU] llama add xpu support #8282

[paddle-bot]

Thanks for your contribution!

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 53.26%. Comparing base (90cef20) to head (c4d5381).
Report is 1 commits behind head on develop.

Additional details and impacted files

@@           Coverage Diff            @@
##           develop    #9152   +/-   ##
========================================
  Coverage    53.26%   53.26%           
========================================
  Files          652      652           
  Lines       105607   105607           
========================================
  Hits         56254    56254           
  Misses       49353    49353           

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