Fix FusedLinearWithGradAdd bug

[MarioLulab]

PR types

Bug fixes

PR changes

Others

Description

llm/gpt3 模型在开启 enable_linear_fused_grad_add 策略后会报错：

具体定位到问题出现在 FusedLinearWithGradAdd 的 backward 方法中。FusedLinearWithGradAdd.backward 会调用 fused_linear_param_grad_add 算子进行计算，而 _C_ops.fused_linear_param_grad_add api 接收六个参数输入：

(Tensor x, Tensor dout, Tensor dweight, Tensor dbias, bool multi_precision = true, bool has_bias = true)

当前 FusedLinearWithGradAdd 在调用 _C_ops.fused_linear_param_grad_add 时缺少第六个参数（has_bias）参数的输入，导致 C++ api 没法正确解析出 has_bias 参数，进而报错

补充说明：
#6602 也修复了类似的问题

[paddle-bot]

Thanks for your contribution!

[ZHUI]

LGTM

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 55.41%. Comparing base (db49062) to head (5742067).
Report is 54 commits behind head on develop.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #8178      +/-   ##
===========================================
- Coverage    56.56%   55.41%   -1.16%     
===========================================
  Files          589      597       +8     
  Lines        89964    91594    +1630     
===========================================
- Hits         50889    50754     -135     
- Misses       39075    40840    +1765     

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