Help Center> ModelArts> Tool Guide> Model Training> Viewing Training Logs

Viewing Training Logs

A training job version is generated each time a training job is executed. The version number is in the format of V0001 and V0002 in ascending order by default. You can view the logs of a training job with different versions.

Output and Log Paths of Training Jobs

When you submit a training job, the system automatically creates a folder with the same name as the training job in the configured OBS path to store the model, logs, and code outputted after training is complete.

For example, when the train-job-01 job is submitted, a folder named train-job-01 is created in the test-modelarts2 bucket. In this folder, three sub-folders (output, log, and code) are created to store the outputted model, logs, and training code, respectively. Sub-folders will be created in the output folder based on your training job version. The following is an example of the folder structure:
test-modelarts2
    |---train-job-01
         |---output
              |---V0001
              |---V0002
              |---...
         |---log
         |---code

Viewing Training Logs in ToolKit

In PyCharm, click ModelArts Training Log in the lower right corner of the page. The training logs are displayed.

Figure 1 Viewing training logs

After the training job is complete, the system automatically creates a folder named MA_LOG in the project directory to store training job log files.

Figure 2 Log files in the project directory

Viewing Training Jobs in OBS

After the training job is complete, the system stores logs to the log directory in an OBS path.

  1. In the ModelArts Explorer area of PyCharm, double-click the version of the training job. The detailed configurations of the training job are displayed in the ModelArts Training Job area.

    You can find the training log output path, that is, the OBS path, in the detailed configurations.

    Figure 3 Finding the log output path
  2. Log in to OBS Console using the current account, find the folder based on the OBS path obtained in the previous step, and view the detailed logs of the training job.