Background

• The one-click model deployment function is only available for datasets whose labeling type is object detection and image classification.
• Currently, only the built-in algorithms can be used for training. You are advised to separately create a training job for a task that is trained using frequently-used frameworks or custom images.
• After a one-click model deployment task is created, the system automatically creates a training job, imports the trained model to ModelArts, and deploys the model as a real-time service. You only need to set parameters once to complete dataset-based AI development.
• Expense description:
  • Training jobs are billed in pay-per-use mode based on the resources you select. Training once is billed one time, and no extra fee is generated.
  • Currently, the imported models are not billed.
  • After the task is created, the created real-time service is in the Running status. If the public resource pool is used, the service is always being billed. Stop the service based on the site requirements to avoid unnecessary fees.
• The OBS directory you use and ModelArts are in the same region.

Creating a Task

1. Log in to the ModelArts management console. In the left navigation pane, choose Data Management > Datasets. The Datasets page is displayed.
2. In the dataset list, choose Deploy Model > Create Task. The Deploy Model page is displayed.
3. Enter a task name and description.
4. Set parameters related to training. See Figure 1.

   When you create a one-click model deployment task, the training parameters are similar to those of the training jobs. You can only select a built-in algorithm for training. A built-in algorithm will be displayed based on the type of your dataset by default. For datasets of the object detection type, the available built-in algorithm is Faster_RCNN_ResNet_v1_50. For datasets of the image classification type, the available built-in algorithm is ResNet_v1_50. Running parameters vary depending on the built-in algorithm. Retain the default values for Running Parameter.

   Set the following parameters: Training Output Path, Job Log Path, Resource Pool, Type, Specifications, and Compute Nodes. For details about how to set the parameters, see Using Built-in Algorithms to Train Models.

   Figure 1 Training information
   
5. Set deployment parameters. See Figure 2.
   Select the resources used for real-time service deployment by setting the following parameters: Resource Pool, Instance Flavor, Instance Count, and Environment Variable. For details about how to set the parameters, go to 3 in Deploying a Model as a Real-Time Service.

   Figure 2 Parameters related to deployment
   
6. Set Sample Collection. This function is disabled by default. To enable this function, configure the related parameters. For details, see Collecting Data.
7. After confirming the entered information, complete the task creation as prompted.

   The task creation process includes the Initialize, Train, Generate Model, and Deploy steps. The time required varies according to dataset size.

   After the Train, Generate Model, and Deploy steps are complete, the View Training Details, View Model Details, and View Service Details links are displayed on the page. You can click these links to view details.

   Figure 3 Task creation process
   

Viewing Task History

1. Log in to the ModelArts management console. In the left navigation pane, choose Data Management > Datasets. The Datasets page is displayed.
2. In the dataset list, choose Deploy Model > View Historical Task. The One-Click Model Deployment Task History page is displayed. Figure 4 shows the details.
   Figure 4 Historical task details