Run fluid models and update current dist train performance result

[typhoonzero]

We need to update performance results under benchmark/cluster to add different models:

codes:

https://github.com/PaddlePaddle/models/tree/develop/fluid

• NLP model:
  • text_classification -- @typhoonzero
  • sequence_tagging_for_ner -- @putcn @helinwang
  • neural_machine_translation --@gongweibao
• image model:
  • image_classification -- @seiriosPlus
  • ResNet 50

Result metrics:

• speed of local mode
• dist mode speed up percentage (CPU/GPU)

[seiriosPlus]

I will add performance about image_classification.

[putcn]

Me and @helinwang will take sequence_tagging_for_ner

[wangkuiyi]

Could we do these performance evaluation via the continuous evaluation system developed by @Superjomn ? If so, it seems that the results will be continuously updated and could be shared publicly.

[typhoonzero]

@wangkuiyi sure! I will move perf tests to CI system once we have the baseline updated.

[Superjomn]

yes, the CE is already online for testing. http://yq01-sys-hic-v100-box-a223-0007.yq01.baidu.com:8111/

One can reference https://github.com/PaddlePaddle/MacroCE/blob/develop/doc/howto.md and PR to https://github.com/Superjomn/paddle-modelci-baseline/tree/fast, the fast branch to add more evaluations. @wangkuiyi @typhoonzero

I will add more docs latter about the branches and how the TeamCity works with it.

[gongweibao]

@wangkuiyi We plan to add later #9046 (comment)

[typhoonzero]

Result google doc link: https://docs.google.com/spreadsheets/d/1D5Xc_TfGfMV5aKh4ZJS_b4js3Mnn06H1Po0iuECZLr4/edit?usp=sharing

[typhoonzero]

@putcn I'm currently working on sequence_tagging_for_ner perf data now

[putcn]

@typhoonzero sorry I was working on aws tool, and thanks for taking this over. I made some progress on ner test, please refer to https://github.com/putcn/Paddle/tree/benchmark_cluster_ner if you need anything.

[typhoonzero]

Thanks! @putcn

[typhoonzero]

Done. Results are in https://docs.google.com/spreadsheets/d/1D5Xc_TfGfMV5aKh4ZJS_b4js3Mnn06H1Po0iuECZLr4/edit?usp=sharing