文档

Migration Operations

  1. Create an Ascend training project.
    1. Click New Project on the welcome page or choose File > New > Project... on the menu bar. The New Project window is displayed. Set corresponding parameters and click Next. See Figure 1.
      • CANN Version: CANN software package version. If the version is not identified or you want to change the version, click Change and select the Ascend-CANN-Toolkit installation path (to the version number or latest level) in the dialog box displayed.
      • Templates: framework project. Select PyTorch Project.
      Figure 1 Project creation
    2. Configure related information and click Finish. See Figure 2.
      • Project name: project name, which can be customized.
      • Project location: project directory. By default, the project is created in $HOME/MindstudioProjects.
      Figure 2 Project configuration
  2. Go to the project page and click (PyTorch GPU2Ascend) on the toolbar.
  3. Go to the PyTorch GPU2Ascend parameter configuration page and set the parameters as shown in Figure 3 (conversion from a single-device GPU to a single-device NPU is used as an example).
    • PyTorch Version: PyTorch version.
    • Input Path: Path of the script file to be migrated.
    • Output Path: Output path of the script after migration.
    • Replace Unsupported APIs: Replace unsupported APIs with APIs with similar functions.
    Figure 3 PyTorch GPU2Ascend parameter settings
  4. After setting the parameters, click Transplant to start the migration.

    If "Transplant success!" is displayed, the migration is successful.

    2022-09-28 12:06:57  [INFO]    [Progress:100.00%]  Package ascend_function has been copy to the output dir, please add /home/username/MindstudioProjects/MyTraining-pytorch/pytorch-cifar100_msft to PYTHONPATH before run net.
2022-09-28 12:06:57  [INFO]    [Progress:100.00%]  MsFmkTransplt run success, welcome to the next use.
2022-09-28 12:06:57  Transplant success!