Platform For AI:Invoke services over the Internet or an internal network through a gateway

1. Get the endpoint and token

After you deploy a service, the system automatically generates the required endpoint and token.

2. Construct and send the request

For both Internet and VPC endpoints, the request construction is the same; only the URL differs. A standard invocation request typically includes four core components:

• Method: The most common methods are POST and GET.

• URL: The base <EAS_ENDPOINT> with the specific Request Path appended.

• Request Header: At a minimum, this must include your authentication token, such as Authorization: <EAS_TOKEN>.

• Request Body: The API of the deployed model determines the format, such as JSON.

  Important

  When invoking a service through a gateway, the request body cannot exceed 1 MB.

Scenario 1: Invoke a model deployed from Model Gallery

Refer directly to the Overview page in the Model Gallery. It provides the most accurate API call examples, typically in cURL or Python, including the complete URL path and Request Body format.

Scenario 2: Invoke a large language model

Large Language Model (LLM) services typically provide OpenAI-compatible API endpoints, such as the chat completions endpoint (/v1/chat/completions) and the completions endpoint (/v1/completions).

For example, to call the chat completions endpoint of a DeepSeek-R1-Distill-Qwen-7B model service deployed with vLLM, you need the following components. For more information, see Invoke an LLM:

• Method: POST

• URI: <EAS_ENDPOINT>/v1/chat/completions

• Request Header: Authorization: <EAS_TOKEN> and Content-Type: application/json

• Request Body:

  {
      "model": "DeepSeek-R1-Distill-Qwen-7B",
      "messages": [
      {
          "role": "system",
          "content": "You are a helpful assistant."
      },
      {
          "role": "user",
          "content": "hello!"
      }
      ]
  }

Example: Invoke with cURL and Python

Assume that <EAS_ENDPOINT> is http://16********.cn-hangzhou.pai-eas.aliyuncs.com/api/predict/test.

curl http://16********.cn-hangzhou.pai-eas.aliyuncs.com/api/predict/test/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: *********5ZTM1ZDczg5OT**********" \
-X POST \
-d '{
    "model": "DeepSeek-R1-Distill-Qwen-7B",
    "messages": [
    {
        "role": "system",
        "content": "You are a helpful assistant."
    },
    {
        "role": "user",
        "content": "hello!"
    }
    ]
}' 

import requests

# Replace with the actual endpoint.
url = 'http://16********.cn-hangzhou.pai-eas.aliyuncs.com/api/predict/test/v1/chat/completions'
# The value of Authorization in the header is the actual token.
headers = {
    "Content-Type": "application/json",
    "Authorization": "*********5ZTM1ZDczg5OT**********",
}
# Construct the service request based on the data format required by the specific model.
data = {
    "model": "DeepSeek-R1-Distill-Qwen-7B",
    "messages": [
        {
            "role": "system",
            "content": "You are a helpful assistant."
        },
        {
            "role": "user",
            "content": "hello!"
        }
    ]
}
# Send the request.
resp = requests.post(url, json=data, headers=headers)
print(resp)
print(resp.content)

More scenarios

• Invoke ComfyUI API.

• Invoke SDWebUI API.

• Invoke a TensorFlow Serving service.

• For services deployed with a universal Processor (including TensorFlow, Caffe, and PMML): See Construct a service request based on a universal processor.

• Models you trained yourself: The invocation method is the same as for the original model.

• Other custom services: The data input format that you define in your custom image or code determines the request format.