Help Center/ ModelArts/ User Guide for Senior AI Engineers (To Be Offline)/ Training Management (Old Version )/ Managing Visualization Jobs
Updated on 2023-10-13 GMT+08:00
View PDF
Share

Managing Visualization Jobs

You can create visualization jobs of TensorBoard and MindInsight types on ModelArts.

TensorBoard and MindInsight can effectively display the change trend of a training job and the data used in the training.

  • TensorBoard

    TensorBoard effectively displays the computational graph of TensorFlow in the running process, the trend of all metrics in time, and the data used in the training. TensorBoard supports only the training jobs based on the TensorFlow or MXNet engine. For more information about TensorBoard, see TensorBoard official website.

  • MindInsight

    MindInsight visualizes information such as scalars, images, computational graphs, and model hyperparameters during training. It also provides functions such as training dashboard, model lineage, data lineage, and performance debugging, helping you train and debug models efficiently. MindInsight supports only the training jobs based on the MindSpore engine. For more information about MindInsight, see MindSpore official website.

You can use the summary file generated during model training to create a visualization job.

Prerequisites

To ensure that the summary file is generated in the training result, add the related code to the training script.

  • Using the TensorFlow engine:

    When using the TensorFlow-based MoXing, in mox.run, set save_summary_steps>0 and summary_verbosity≥1.

    If you want to display other metrics, add tensors to log_info in the return value mox.ModelSpec of model_fn. Only the rank-0 tensors (scalars) are supported. The added tensors are written into the summary file. If you want to write tensors of higher ranks in the summary file, use the native tf.summary of TensorFlow in model_fn.

  • Using the MindSpore engine:

    MindSpore allows you to save data to the summary log file and display the data on the GUI. For details, see Collecting Summary Record.

  • Using the MXNet engine:

    Add the following code to the script:

    batch_end_callbacks.append(mx.contrib.tensorboard.LogMetricsCallback('OBS path'))

Precautions

  • You will be charged as long as your visualization jobs are in the Running status. We recommend you to stop the visualization jobs when you no longer need them to prevent unnecessary fees. Visualization jobs can be automatically stopped at the specified time. To prevent unnecessary fees, enable this function.
  • By default, CPU resources are used to run visualization jobs and cannot be changed to other resources.
  • The OBS directory you use and ModelArts are in the same region.

Creating a Visualization Job

  1. Log in to the ModelArts management console. In the left navigation pane, choose Training Jobs. On the displayed page, click the Visualization Jobs tab.
  2. In the upper left corner of the visualization job list, click Create to switch to the Create Visualization Job page.
  3. Set Billing Mode to Pay-per-use and Job Type to TensorBoard and MindInsight. Enter the visualization job name and description as required, set the Training Output Path and Auto Stop parameters.
    • Training Output Path: Select the training output path specified when the training job is created.
    • Auto Stop: Enable or disable the auto stop function. A running visualization job will be billed. To prevent unnecessary fees, enable the auto stop function to automatically stop the visualization job at the specified time. The options are 1 hour later, 2 hours later, 4 hours later, 6 hours later, and Custom. If you select Custom, you can enter any integer within 1 to 24 hours in the text box on the right.
    Figure 1 Creating a visualization job
  4. Click Next.
  5. After confirming the specifications, click Next.

    In the visualization job list, when the status changes to Running, the virtualization job has been created. You can click the name of the visualization job to view its details.

Opening a Visualization Job

In the visualization job list, click the name of the target visualization job. The page is displayed. Only the visualization job in the Running status can be opened.

Figure 2 TensorBoard page
Figure 3 MindInsight GUI

Running or Stopping a Visualization Job

  • Stopping a visualization job: Stop a running visualization job when it is no longer needed to stop billing. In the visualization job list, click Stop in the Operation column to stop the visualization job.
  • Running a visualization job: You can run and use a visualization job in the Canceled status again. In the visualization job list, click Run in the Operation column to run the visualization job.

Deleting a Visualization Job

If your visualization job is no longer used, you can delete it to release resources. In the visualization job list, click Delete in the Operation column to delete the visualization job.

A deleted visualized job cannot be recovered. Create a new visualization job if you want to use it. Exercise caution when performing this operation.