Procedure

Preparations

• Download the ONNX model tuning repository.
• Prepare the ONNX model file (.onnx) to be tuned.

Procedure

To configure the one-click performance tuning tool, perform the following steps:

1. Select and open an application project.
2. Choose Ascend > Advisor > Advisor Pro from the menu bar. The corresponding tool page is displayed. See Figure 1.
   Figure 1 One-click performance tuning page
   
3. Select Inference Optimization on the Figure 1. The configuration page is displayed, on which you can select Remote Run or Local Run. Figure 2 uses Remote Run as an example.
   Figure 2 Remote Run
   

   Table 1 Parameters

   Parameter	Description
   Run Mode	Remote Run Local Run
   Deployment	Run configuration. This parameter is mandatory and is available only when Remote Run is selected. You can use the Deployment function to synchronize the files and folders in a specified project to a specified directory on a remote device. For details, see Ascend Deployment.
   Project Location	Directory for storing the .json file after analysis. The default value is ${HOME}/AscendProjects/AscendAdvisor/untitled. This parameter is mandatory and can be customized.
   ONNX Location	ONNX model directory. Specify this parameter to a file suffixed with .onnx.
   Environment Variables	Environment variable configuration. You need to configure the environment variables on which Python depends. PYTHONPATH=${PATH}/msadvisor-master:$PYTHONPATH LD_LIBRARY_PATH=${HOME}/python3/lib:$LD_LIBRARY_PATH ${PATH} indicates the directory for storing the ONNX model tuning repository, and ${HOME} indicates the home directory for installing Python.
   Remote CANN Path	Installation path of the CANN package in the remote operating environment, available when you select the Remote Run mode. This parameter is mandatory. For example, set this parameter to Ascend-cann-toolkit_installation_path/ascend-toolkit/{version}.
   KnowledgeBase Configuration	Repository configuration file ecosystem.json. The file is stored in the ONNX model tuning repository's directory, for example, ${PATH}/msadvisor-master/ecosystem/onnx_refactor/ecosystem.json.

4. After the configuration is complete, click Start to start the performance bottleneck analysis by Advisor.

   Figure 3 shows the analysis process. You can click Cancel to cancel the task or click Previous to reconfigure parameters.

   Figure 3 Analysis process
   

   Figure 4 shows the analysis result.

   Figure 4 Analysis result
   
   • Items marked with do not need to be tuned.
   • For items marked with , select each item and click Optimization to perform one-click performance tuning. You can click See More... to view details about the tuning item, as shown in Figure 5.
   Figure 5 See More
   
5. Click Optimization to perform tuning. Figure 6 shows the result.
   Figure 6 Tuning result
   

   The one-click performance tuning capability is being improved.