From bf13c06e1e7b92736caa9310265dce6e9ec8e2d7 Mon Sep 17 00:00:00 2001 From: wangyue01 Date: Tue, 23 Jun 2020 11:15:56 +0800 Subject: [PATCH] Add some docs for Profiler --- README.md | 8 ++++++++ mindinsight/profiler/README.md | 6 ++++-- 2 files changed, 12 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index fa91b3a..52c504c 100644 --- a/README.md +++ b/README.md @@ -11,6 +11,10 @@ such as computation graph, training process metrics, etc. Provide visualization of model parameters information, such as training data, model accuracy, etc. +- Visualization of training performance: + + Provide visualization of training performance information, such as operator execution time, +data input pipeline performance, etc. # Index @@ -91,6 +95,10 @@ The GUI of MindInsight displays the pipeline of dataset processing and augmentat The GUI of MindInsight displays the parameters and operations of the dataset processing and augmentation. +### Performance visualization + +The GUI of MindInsight displays the performance data of the neural networks. + # Installation See [Install MindInsight](https://www.mindspore.cn/install/en). diff --git a/mindinsight/profiler/README.md b/mindinsight/profiler/README.md index 788d5ae..0a57b6c 100644 --- a/mindinsight/profiler/README.md +++ b/mindinsight/profiler/README.md @@ -65,8 +65,8 @@ Figure 2 displays the Step Trace page. The Step Trace detail will show the start can also choose a specific step to see its step trace statistics. The graphs at the bottom of the page show how the execution time of Step Gap, Forward/Backward Propagation and Step Tail changes according to different steps, it will help to decide whether we can optimize the performance of some stages. -*Notice:* MindSpore choose the Foward Start/Backward End Operators automatically, The names of the two operators are shown on the page. It is possible that the two operators are -not choosen as what the user expect. Users can choose the operators from the dumped execution graph, and specify the two operators manually by setting the `FP_POINT` and `BP_POINT` environment. +*Notice:* MindSpore choose the Foward Start/Backward End Operators automatically, The names of the two operators are shown on the page. Profiler do not guarantee that the two operators are +always chosen as the user's expectation. Users can choose the two operators according to the execution graph, and specify the them manually by setting the `FP_POINT` and `BP_POINT` environment variables. For example: `export FP_POINT=fp32_vars/conv2d/conv2Dfp32_vars/BatchNorm/FusedBatchNorm_Reduce` and `export BP_POINT=loss_scale/gradients/AddN_70`. ### Operator Performance Analysis @@ -160,3 +160,5 @@ The Profiler has the following limitations now: * Only programs running on Ascend chip is supported. * To limit the data size generated by the Profiler, MindInsight suggests that for large neural network, the profiled steps should better below 10. +* The parse of Timeline data is time consuming, and several step's data is usually enough for analysis. In order to speed up the data parse and UI +display, Profiler will show at most 20M data (Contain 10+ step information for large networks). \ No newline at end of file -- GitLab