Add global param to refined_recompute

[DongBaiYue]

PR types

New features

PR changes

APIs

Description

Add global param to refined_recompute
不同设备的显存大小可能不同，添加global参数可以精细调节重计算百分比，方便在不同设备上达到显存占用和性能的平衡。
refined_recompute: "global:x"的含义：n层ErnieDecoderLayer中有x层不重计算，n-x层重计算。默认值为0。
n由模型结构确定，以ERNIE VL 28B为例，此时n=14。

• A100（80GB）上可以默认所有层都进行重计算。
• P800（96GB）上可以设置global:3，以关闭3/14的重计算，换取性能提升。

[paddle-bot]

Thanks for your contribution!

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (develop@344c970). Learn more about missing BASE report.

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #2899   +/-   ##
==========================================
  Coverage           ?   31.33%           
==========================================
  Files              ?      420           
  Lines              ?    68486           
  Branches           ?        0           
==========================================
  Hits               ?    21458           
  Misses             ?    47028           
  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.

[DongBaiYue]

/re-run all-failed

[yongqiangma]

LGTM