Help Center> ModelArts> Troubleshooting> Training Jobs> In-Cloud Migration Adaptation Issues> Error Message "No module named .*" Displayed in Training Job Logs
Updated on 2023-01-17 GMT+08:00

Error Message "No module named .*" Displayed in Training Job Logs

Perform the following operations to locate the fault:

  1. Checking Whether the Dependency Package Is Available
  2. Checking Whether the Dependency Package Path Can Be Detected
  3. Checking Whether the Selected Resource Flavor Is Correct
  4. Summary and Suggestions

Checking Whether the Dependency Package Is Available

If the dependency package is unavailable, use either of the following methods to install it:

  • Method 1 (recommended): When you create an algorithm, place the required file or installation package in the code directory.

    The required file varies depending on the dependency package type.

    • If the dependency package is an open-source installation package

      Create a file named pip-requirements.txt in the code directory, and specify the dependency package name and version in the format of Package name==Version in the file.

      For example, the OBS path specified by Code Directory contains model files and the pip-requirements.txt file. The code directory structure is as follows:

      |---OBS path to the model boot file
           |---model.py               # Model boot file
           |---pip-requirements.txt   # Customized configuration file, which specifies the name and version of the dependency package

      The following shows the content of the pip-requirements.txt file:

      alembic==0.8.6
      bleach==1.4.3
      click==6.6
    • If the dependency package is a WHL package

      If the training backend does not support the download of open-source installation packages or the use of custom WHL packages, the system cannot automatically download and install the package. In this case, place the WHL package in the code directory, create a file named pip-requirements.txt, and specify the name of the WHL package in the file. The dependency package must be in WHL format.

      For example, the OBS path specified by Code Directory contains model files, the WHL file, and the pip-requirements.txt file. The code directory structure is as follows:

      |---OBS path to the model boot file
           |---model.py               # Model boot file
           |---XXX.whl                # Dependency package. If multiple dependencies are required, place all of them here.
           |---pip-requirements.txt   # Customized configuration file, which specifies the name of the dependency package

      The following shows the content of the pip-requirements.txt file:

      numpy-1.15.4-cp36-cp36m-manylinux1_x86_64.whl
      tensorflow-1.8.0-cp36-cp36m-manylinux1_x86_64.whl
  • Method 2: Add the following code to the boot file to install the dependency package:
    import os
    os.system('pip install xxx')

In method 1, the dependency package can be downloaded and installed before the training job is started. In method 2, the dependency package is downloaded and installed during the running of the boot file.

Checking Whether the Dependency Package Path Can Be Detected

Before executing code locally, add project_dir to PYTHONPATH or install project_dir in site-package. ModelArts enables you to add project_dir to sys.path to resolve this issue.

Run from module_dir import module_file to import a package. The code structure is as follows:

project_dir
|- main.py
|- module_dir
|  |- __init__.py
|  |- module_file.py

Checking Whether the Selected Resource Flavor Is Correct

Error message "No module named npu_bridge.npu_init" is displayed for a training job.

from npu_bridge.npu_init import *
ImportError: No module named npu_bridge.npu_init

Check whether the flavor used by the training job supports NPUs. The possible cause is that the job selected a non-NPU flavor, for example, a GPU flavor. As a result, an error occurs when NPUs are used.

Summary and Suggestions

Before creating a training job, use the ModelArts development environment to debug the training code to maximally eliminate errors in code migration.