[AutoParallel] Add test for PIR refined recompute

[waliwali777]

PR types

Others

PR changes

Others

Description

Judge refined_ops_patterns training args.
Add test for PIR refined recompute.

在 yaml 文件中的调用举例：

 enable_auto_parallel: 1
 to_static: 1
 recompute: 1
 refined_ops_patterns:
        - {main_ops: [flash_attn], num: -1, pre_ops: [], suf_ops: []}
        - {main_ops: [matmul], num: 2, pre_ops: [], suf_ops: [add]}

在python命令行参数中的调用举例

python -u -m paddle.distributed.launch \
                    --gpus "0,1,2,3" \
                    --log_dir $case_log_dir \
                    run_pretrain_auto.py \
                    --to_static \
                    --enable_auto_parallel 1 \ 
                    --recompute 1 \
                    --refined_ops_patterns '[{"main_ops":["matmul"],"num":-1,"pre_ops":["softmax"],"suf_ops":[]}]' \
                   ...

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Attention: Patch coverage is 40.00000% with 6 lines in your changes missing coverage. Please review.

Project coverage is 52.49%. Comparing base (a52035f) to head (57da8d7).
Report is 3 commits behind head on develop.

Files with missing lines	Patch %	Lines
paddlenlp/trainer/auto_training_args.py	33.33%	6 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #9679      +/-   ##
===========================================
+ Coverage    52.46%   52.49%   +0.03%     
===========================================
  Files          727      723       -4     
  Lines       115028   114327     -701     
===========================================
- Hits         60353    60020     -333     
+ Misses       54675    54307     -368     

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

[ZHUI]

如果这个暂时不能合并的话，要不先放到 paddlenlp/trainer/auto_training_args.py 中？

[waliwali777]

done

[ZHUI]

LGTM