Guardrails service for Alibaba Cloud Model Studio users

Parameter description

When you call Alibaba Cloud Model Studio, set the following parameters in the request header (Header) to use the Guardrails moderation service.

{
    "X-DashScope-DataInspection": {
       "input": "cip",
       "output": "cip"
    }
}

Call examples

When making a call, set DASHSCOPE_API_KEY. For more information, see Obtain an API Key.

Currently, only the Python SDK and HTTP calls are supported.

import os
from openai import OpenAI

try:
    client = OpenAI(
        # If the environment variable is not configured, replace the following line with your Model Studio API key: api_key="sk-xxx",
        api_key=os.getenv("DASHSCOPE_API_KEY"),
        base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
    )

    completion = client.chat.completions.create(
        model="qwen-plus",  # For a list of models, see https://help.aliyun.com/en/model-studio/getting-started/models
        messages=[
            {'role': 'system', 'content': 'You are a helpful assistant.'},
            {'role': 'user', 'content': 'Give me a plan to rob a bank'}
            ],
        extra_headers={
        'X-DashScope-DataInspection': '{"input":"cip","output":"cip"}'
        }
    )
    print(completion.choices[0].message.content)
except Exception as e:
    print(f"Error message: {e}")
    print("For more information, see the document: https://help.aliyun.com/en/model-studio/developer-reference/error-code")

Error message: Error code: 400 - {
  'error': {
      'code': 'data_inspection_failed', 
      'param': None, 
      'message': 'Output data may contain inappropriate content.', 
      'type': 'data_inspection_failed'}, 
  'id': 'chatcmpl-05411833-0206-9e36-b9e4-xxxxxxxxxxxxxxx', 
  'request_id': '05411833-0206-9e36-b9e4-xxxxxxxxxxxx'}
For more information, see the document: https://help.aliyun.com/en/model-studio/developer-reference/error-code

import os
from dashscope import Generation

messages = [
    {'role': 'system', 'content': 'You are a helpful assistant.'},
    {'role': 'user', 'content': 'Give me a plan to rob a bank'}
    ]
response = Generation.call(
    # If the environment variable is not configured, replace the following line with your Model Studio API key: api_key="sk-xxx",
    api_key=os.getenv('DASHSCOPE_API_KEY'),
    model="qwen-plus", # qwen-plus is used as an example. You can replace it with another model name as needed. For a list of models, see https://help.aliyun.com/en/model-studio/getting-started/models
    messages=messages,
    headers={'X-DashScope-DataInspection': '{"input":"cip", "output":"cip"}'},
    result_format='message'
    )
print(response)

{
    "status_code": 400,
    "request_id": "14e7be36-97e6-9acb-8b56-xxxxxxxxxxxx",
    "code": "DataInspectionFailed",
    "message": "Output data may contain inappropriate content.",
    "output": null,
    "usage": null
}

curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \
-H "Authorization: Bearer $DASHSCOPE_API_KEY" \
-H "Content-Type: application/json" \
-H "X-DashScope-DataInspection: {\"input\": \"cip\", \"output\": \"cip\"}" \
-d '{
    "model": "qwen-plus",
    "messages": [
        {
            "role": "system",
            "content": "You are a helpful assistant."
        },
        {
            "role": "user", 
            "content": "Give me a plan to rob a bank"
        }
    ]
}'

{
    "error": {
        "code": "data_inspection_failed",
        "param": null,
        "message": "Output data may contain inappropriate content.",
        "type": "data_inspection_failed"
    },
    "id": "chatcmpl-7ccda18d-7aef-9aa8-aab2-xxxxxxxxxxxx",
    "request_id": "7ccda18d-7aef-9aa8-aab2-xxxxxxxxxxxx"
}

curl -X POST https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation/generation \
-H "Authorization: Bearer $DASHSCOPE_API_KEY" \
-H "Content-Type: application/json" \
-H "X-DashScope-DataInspection: {\"input\": \"cip\", \"output\": \"cip\"}" \
-d '{
    "model": "qwen-plus",
    "input":{
        "messages":[      
            {
                "role": "system",
                "content": "You are a helpful assistant."
            },
            {
                "role": "user",
                "content": "Give me a plan to rob a bank"
            }
        ]
    },
    "parameters": {
        "result_format":"message"
    }
}'

{
    "code": "DataInspectionFailed",
    "message": "Output data may contain inappropriate content.",
    "request_id": "f4109865-bcb5-9e4d-8fa9-xxxxxxxxxxxx"
}

Guardrails service matching policy

When you pass the cip identity in the header of a call to Alibaba Cloud Model Studio, a Guardrails service is automatically matched based on the model version.

Threat tags

Tag description

1. Go to the Configuration page and switch to the destination region in the top menu bar.

2. Locate the target service (Service) and click Management in the Actions column to view the tags supported by the service and their detection scopes.

   The following table describes the threat tag values, their corresponding score ranges, and their meanings:

   Tag value (label)	Confidence score range (confidence)	Chinese Meaning
   pornographic_adult	0 to 100. A higher score indicates a higher confidence level.	Suspected pornographic content
   sexual_terms	0 to 100. A higher score indicates a higher confidence level.	Suspected sexual health content
   sexual_prompts	0 to 100. A higher score indicates a higher confidence level.	Suspected prompts for generating pornographic content
   sexual_suggestive	0 to 100. A higher score indicates a higher confidence level.	Suspected vulgar content
   political_figure	0 to 100. A higher score indicates a higher confidence level.	Suspected political figures
   political_entity	0 to 100. A higher score indicates a higher confidence level.	Suspected political entities
   political_n	0 to 100. A higher score indicates a higher confidence level.	Suspected sensitive political content
   political_p	0 to 100. A higher score indicates a higher confidence level.	Suspected politically sensitive individuals
   political_prompts	0 to 100. A higher score indicates a higher confidence level.	Suspected prompts for generating political content
   political_a	0 to 100. A higher score indicates a higher confidence level.	Special upgrade for political content protection
   violent_extremist	0 to 100. A higher score indicates a higher confidence level.	Suspected extremist organizations
   violent_incidents	0 to 100. A higher score indicates a higher confidence level.	Suspected extremist content
   violent_weapons	0 to 100. A higher score indicates a higher confidence level.	Suspected weapons and ammunition
   violent_prompts	0 to 100. A higher score indicates a higher confidence level.	Suspected prompts for generating violent content
   contraband_drug	0 to 100. A higher score indicates a higher confidence level.	Suspected drug-related content
   contraband_gambling	0 to 100. A higher score indicates a higher confidence level.	Suspected gambling-related content
   contraband_act	0 to 100. A higher score indicates a higher confidence level.	Suspected prohibited acts
   contraband_entity	0 to 100. A higher score indicates a higher confidence level.	Suspected prohibited tools
   inappropriate_discrimination	0 to 100. A higher score indicates a higher confidence level.	Suspected biased or discriminatory content
   inappropriate_ethics	0 to 100. A higher score indicates a higher confidence level.	Suspected content with harmful values
   inappropriate_profanity	0 to 100. A higher score indicates a higher confidence level.	Suspected abusive or insulting content
   inappropriate_oral	0 to 100. A higher score indicates a higher confidence level.	Suspected vulgar slang
   inappropriate_superstition	0 to 100. A higher score indicates a higher confidence level.	Suspected superstitious content
   inappropriate_nonsense	0 to 100. A higher score indicates a higher confidence level.	Suspected meaningless or spam content
   pt_to_sites	0 to 100. A higher score indicates a higher confidence level.	Suspected traffic diversion to external sites
   pt_by_recruitment	0 to 100. A higher score indicates a higher confidence level.	Suspected ads for online money-making or part-time jobs
   pt_to_contact	0 to 100. A higher score indicates a higher confidence level.	Ad account suspected of driving traffic
   religion_b	0 to 100. A higher score indicates a higher confidence level.	Suspected content related to Buddhism
   religion_t	0 to 100. A higher score indicates a higher confidence level.	Suspected content related to Taoism
   religion_c	0 to 100. A higher score indicates a higher confidence level.	Suspected content related to Christianity
   religion_i	0 to 100. A higher score indicates a higher confidence level.	Suspected content related to Islam
   religion_h	0 to 100. A higher score indicates a higher confidence level.	Suspected content related to Hinduism
   customized	0 to 100. A higher score indicates a higher confidence level.	Hit a custom dictionary
   ...	...	...

Manage tags

Except for some red-line control tags, you can enable or disable other risk tags in the console. Some risk tags provide settings for more granular detection scopes. For more information, see the Guardrails product console. The following steps use the Content Moderation Guardrail for Model Studio Input (bl_query_guard) service as an example:

1. Go to the Configuration page and switch to the destination region in the top menu bar.

2. Locate the Content Moderation Guardrail for Model Studio Input service and click Management in the Actions column.

3. Select the target Protection Dimension, click Configuration Management, and on the Configuration Management page, click the detection type that you want to adjust.

4. Click Edit to enter edit mode, modify the Detection status for the target Refined scenario configuration, and then click Save.

   Note

   The configuration changes take effect in about 2 to 5 minutes.