After you deploy a service in Platform for AI (PAI), you can use the online debugging feature to test whether the service runs as expected. This topic describes how to debug a service online.
Procedure
Log on to the PAI console. Select a region on the top of the page. Then, select the desired workspace and click Enter Elastic Algorithm Service (EAS).
On the Inference Service tab, find the service that you want to manage, click
in the Actions column, and then select Online Debugging.In the Request Parameter Online Tuning section of the Online Debugging tab, configure request parameters and click Send Request.
In this example, the vLLM-based Qwen2.5-7B-Instruct model service is used. You need to add
/v1/chat/completionsto the existing URL to debug the service.
Different model services may differ in their request methods, URLs, and request bodies. Configure request data based on your model service.
If you deploy a large language model (LLM) service, see Online debugging.
If you deploy a ComfyUI model service, see Call an EAS service by using API operations.
If you deploy a model service by using a common processor, such as TensorFlow, Caffe, or Predictive Model Markup Language (PMML), see Construct requests for services based on a universal processor.
If you deploy a model service in Model Gallery, see the content on the Overview tab of the model details page to debug your service.
If you deploy other types of model services, configure the request data based on the input data format of your custom models or images.
References
You can use an automatic stress testing tool to create stress testing tasks for services that are deployed in Elastic Algorithm Service (EAS) to learn more about the performance of EAS. For more information, see Automatic service stress testing.
For information about EAS use cases, see EAS use cases.