After a service is deployed, you can use the online debugging feature in the console to test whether the service runs as expected. This feature provides a graphical interface where you can send test requests and view the results without writing any code.
How to debug online
Go to the online debugging page.
Log on to the PAI console. Select a region on the top of the page. Then, select the desired workspace and click Elastic Algorithm Service (EAS).
On the Inference Service tab, find the service that you want to test and click
> Online Debugging in the Actions column.
Construct and send the request (key step). The definitions for API operations, especially the URI of the request and the request body format, vary significantly depending on the model. Ensure that you enter the request parameters correctly. A standard request includes the following elements:
Method: Usually
POSTorGET.URI of the request (URL): The base address is automatically populated for online debugging. You must determine whether to append a specific API operation path. An incorrect path is the most common cause of a
404 Not Founderror.Request header (Headers): The
Authorizationtoken is usually automatically populated and does not need to be modified.Request body (Body): The request body must comply with the API operation requirements and be in the correct format. An incorrect format is the primary cause of a
400 Bad Requestor500 Internal Server Errorerror.
Usage example
This example shows how to test the chat API operation for the Qwen2.5-7B-Instruct model service deployed using vLLM. The parameters are as follows:
Method: POST
URI of the request (URL):
http://***********/v1/chat/completions(You need to append/v1/chat/completionsto the URL that is automatically populated for online debugging.)Request body (Body):
{ "model": "Qwen2.5-7B-Instruct", "messages": [ { "role": "user", "content": "What is the captial of Canada?" } ] }
The result is as follows:
FAQ
Q: How do I determine the correct URL and request body?
The correct URL and request body depend on the model that you deployed. For more information, see Construct requests for typical scenarios.
Next steps
Performance evaluation: You can use the automatic service stress testing feature to evaluate the performance of your service.
Official invocation: After the service performance meets your requirements, you can invoke the service. For more information, see Service invocation methods.