Deploying a Model

You can deploy a model as a real-time service that provides a real-time test UI and monitoring capabilities. After model training is complete, you can deploy a version with the ideal accuracy and in the Successful status as a service. The procedure is as follows:

1. On the Train Model tab page, wait until the training status changes to Successful. Click Deploy in the Version Manager pane to deploy the model as a real-time service.
   Figure 1 Deployment operation
   
2. In the Deploy dialog box, select resource flavor, set the Auto Stop function, and click OK to start the deployment.
   • Specifications: Available specifications are Free (CPU), Free (GPU), Compute-intensive 3 instance (CPU), and Compute-intensive 2 instance (GPU). The GPU specifications are better, and the CPU specifications are more cost-effective. If you want to use free specifications, read the information displayed on the page.
   • Compute Nodes: The default value is 1 and cannot be changed.
   • Auto Stop: After this parameter is enabled and the auto stop time is set, a service automatically stops at the specified time. If this parameter is disabled, a real-time service is always running and billed. The function can help you avoid unnecessary charges. The auto stop function is enabled by default. The default value is 1 hour later.

   Currently, 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 from 1 to 24 hours in the text box on the right.

   Figure 2 Deploying a model
   
3. After the model is deployed, view the model deployment status on the Service Deployment page.

   The deployment takes a certain period of time. If the status in the Version Manager pane changes from Deploying to Running, the deployment is complete.

   

   On the ExeML page, trained models can only be deployed as real-time services. For details about how to deploy them as batch services or edge services, see Where Are Models Generated by ExeML Stored? What Other Operations Are Supported?.

Testing a Service

• On the Service Deployment page, select a service type. For example, on the ExeML page, the text classification model is deployed as a real-time service by default. On the Real-Time Services page, click Prediction in the Operation column of the target service to perform a service test. For details, see Testing a Service.
• You can also use code to test a service. For details, see Accessing a Real-Time Service and Accessing an Edge Service.
• The following describes the procedure for performing a service test after the text classification model is deployed as a service on the ExeML page.
  1. After the model is deployed, you can add a text file for test. On the ExeML page, click the target project, go to the Deploy Service tab page, select the service version in the Running status, and enter text to be tested in the text box in the Service Test area.
  2. Click Prediction to perform the test. After the prediction is complete, the result is displayed in the Prediction Result area on the right. If the model accuracy does not meet your expectation, add text data files on the Label Data tab page, label the files, and train and deploy the model again. Table 1 describes the parameters in the prediction result. If you are satisfied with the model prediction result, call the API to access the real-time service as prompted. For details, see Accessing a Real-Time Service.
     Figure 3 Prediction
     

     Table 1 Parameters in the prediction result

     Parameter	Description
     predicted_label	Prediction type of the text
     score	Confidence score for the predicated class

     

     A running real-time service keeps consuming the resources. If you do not need to use the real-time service, you are advised to click Stop in the Version Manager pane to stop the service and avoid unnecessary charges. If you want to use the service again, click Start.

     If you enable the auto stop function, the service automatically stops after the specified time and no fee is generated.