敏感内容审查

除了系统内置的内容审查类型,Dify 也支持用户扩展自定义的内容审查规则,该方法适用于私有部署的开发者定制开发。比如企业内部客服,规定用户在查询的时候以及客服回复的时候,除了不可以输入暴力,性和非法行为等相关词语,也不能出现企业自己规定的禁词或违反内部制定的审查逻辑,那么开发者可以在私有部署的 Dify 代码层扩展自定义内容审查规则。

快速开始

这里以一个 Cloud Service 内容审查扩展为例,步骤如下:

  1. 初始化目录

  2. 添加前端组件定义文件

  3. 添加实现类

  4. 预览前端界面

  5. 调试扩展

1. 初始化目录

新增自定义类型 Cloud Service,需要在 api/core/moderation 目录下新建相关的目录和文件。

.
└── api
    └── core
        └── moderation
            └── cloud_service
                ├── __init__.py
                ├── cloud_service.py
                └── schema.json

2.添加前端组件规范

{
    "label": {
        "en-US": "Cloud Service",
        "zh-Hans": "云服务"
    },
    "form_schema": [
        {
            "type": "select",
            "label": {
                "en-US": "Cloud Provider",
                "zh-Hans": "云厂商"
            },
            "variable": "cloud_provider",
            "required": true,
            "options": [
                {
                    "label": {
                        "en-US": "AWS",
                        "zh-Hans": "亚马逊"
                    },
                    "value": "AWS"
                },
                {
                    "label": {
                        "en-US": "Google Cloud",
                        "zh-Hans": "谷歌云"
                    },
                    "value": "GoogleCloud"
                },
                {
                    "label": {
                        "en-US": "Azure Cloud",
                        "zh-Hans": "微软云"
                    },
                    "value": "Azure"
                }
            ],
            "default": "GoogleCloud",
            "placeholder": ""
        },
        {
            "type": "text-input",
            "label": {
                "en-US": "API Endpoint",
                "zh-Hans": "API Endpoint"
            },
            "variable": "api_endpoint",
            "required": true,
            "max_length": 100,
            "default": "",
            "placeholder": "https://api.example.com"
        },
        {
            "type": "paragraph",
            "label": {
                "en-US": "API Key",
                "zh-Hans": "API Key"
            },
            "variable": "api_keys",
            "required": true,
            "default": "",
            "placeholder": "Paste your API key here"
        }
    ]
}

3. 添加实现类

cloud_service.py 代码模版,你可以在这里实现具体的业务逻辑。

注意:类变量 name 为自定义类型名称,需要跟目录和文件名保持一致,而且唯一。

from core.moderation.base import Moderation, ModerationAction, ModerationInputsResult, ModerationOutputsResult

class CloudServiceModeration(Moderation):
    """
    The name of custom type must be unique, keep the same with directory and file name.
    """
    name: str = "cloud_service"

    @classmethod
    def validate_config(cls, tenant_id: str, config: dict) -> None:
        """
        schema.json validation. It will be called when user save the config.

        Example:
            .. code-block:: python
                config = {
                    "cloud_provider": "GoogleCloud",
                    "api_endpoint": "https://api.example.com",
                    "api_keys": "123456",
                    "inputs_config": {
                        "enabled": True,
                        "preset_response": "Your content violates our usage policy. Please revise and try again."
                    },
                    "outputs_config": {
                        "enabled": True,
                        "preset_response": "Your content violates our usage policy. Please revise and try again."
                    }
                }

        :param tenant_id: the id of workspace
        :param config: the variables of form config
        :return:
        """

        cls._validate_inputs_and_outputs_config(config, True)

        if not config.get("cloud_provider"):
            raise ValueError("cloud_provider is required")

        if not config.get("api_endpoint"):
            raise ValueError("api_endpoint is required")

        if not config.get("api_keys"):
            raise ValueError("api_keys is required")

    def moderation_for_inputs(self, inputs: dict, query: str = "") -> ModerationInputsResult:
        """
        Moderation for inputs.

        :param inputs: user inputs
        :param query: the query of chat app, there is empty if is completion app
        :return: the moderation result
        """
        flagged = False
        preset_response = ""

        if self.config['inputs_config']['enabled']:
            preset_response = self.config['inputs_config']['preset_response']

            if query:
                inputs['query__'] = query
            flagged = self._is_violated(inputs)

        # return ModerationInputsResult(flagged=flagged, action=ModerationAction.overridden, inputs=inputs, query=query)
        return ModerationInputsResult(flagged=flagged, action=ModerationAction.DIRECT_OUTPUT, preset_response=preset_response)

    def moderation_for_outputs(self, text: str) -> ModerationOutputsResult:
        """
        Moderation for outputs.

        :param text: the text of LLM response
        :return: the moderation result
        """
        flagged = False
        preset_response = ""

        if self.config['outputs_config']['enabled']:
            preset_response = self.config['outputs_config']['preset_response']

            flagged = self._is_violated({'text': text})

        # return ModerationOutputsResult(flagged=flagged, action=ModerationAction.overridden, text=text)
        return ModerationOutputsResult(flagged=flagged, action=ModerationAction.DIRECT_OUTPUT, preset_response=preset_response)

    def _is_violated(self, inputs: dict):
        """
        The main logic of moderation.

        :param inputs:
        :return: the moderation result
        """
        return False

4. 调试扩展

至此,即可在 Dify 应用编排界面选择自定义的 Cloud Service 内容审查扩展类型进行调试。\

实现类模版

from core.moderation.base import Moderation, ModerationAction, ModerationInputsResult, ModerationOutputsResult

class CloudServiceModeration(Moderation):
    """
    The name of custom type must be unique, keep the same with directory and file name.
    """
    name: str = "cloud_service"

    @classmethod
    def validate_config(cls, tenant_id: str, config: dict) -> None:
        """
        schema.json validation. It will be called when user save the config.
        
        :param tenant_id: the id of workspace
        :param config: the variables of form config
        :return:
        """
        cls._validate_inputs_and_outputs_config(config, True)
        
        # implement your own logic here

    def moderation_for_inputs(self, inputs: dict, query: str = "") -> ModerationInputsResult:
        """
        Moderation for inputs.

        :param inputs: user inputs
        :param query: the query of chat app, there is empty if is completion app
        :return: the moderation result
        """
        flagged = False
        preset_response = ""
        
        # implement your own logic here
        
        # return ModerationInputsResult(flagged=flagged, action=ModerationAction.overridden, inputs=inputs, query=query)
        return ModerationInputsResult(flagged=flagged, action=ModerationAction.DIRECT_OUTPUT, preset_response=preset_response)

    def moderation_for_outputs(self, text: str) -> ModerationOutputsResult:
        """
        Moderation for outputs.

        :param text: the text of LLM response
        :return: the moderation result
        """
        flagged = False
        preset_response = ""
        
        # implement your own logic here

        # return ModerationOutputsResult(flagged=flagged, action=ModerationAction.overridden, text=text)
        return ModerationOutputsResult(flagged=flagged, action=ModerationAction.DIRECT_OUTPUT, preset_response=preset_response)

实现类开发详细介绍

def validate_config

schema.json 表单校验方法,当用户点击「发布」保存配置时调用

  • config 表单参数

    • {{variable}} 表单自定义变量

    • inputs_config 输入审查预设回复

      • enabled 是否开启

      • preset_response 输入预设回复

    • outputs_config输出审查预设回复

      • enabled 是否开启

      • preset_response 输出预设回复

def moderation_for_inputs

输入校验函数

  • inputs :终端用户传入变量值

  • query :终端用户当前对话输入内容,对话型应用固定参数。

  • ModerationInputsResult

    • flagged 是否违反校验规则

    • action 执行动作

      • direct_output 直接输出预设回复

      • overridden 覆写传入变量值

    • preset_response 预设回复(仅当 action=direct_output 返回)

    • inputs 终端用户传入变量值,key 为变量名,value 为变量值(仅当 action=overridden 返回)

    • query 覆写的终端用户当前对话输入内容,对话型应用固定参数。(仅当 action=overridden 返回)

def moderation_for_outputs

输出校验函数

  • text :模型输出内容

  • moderation_for_outputs :输出校验函数

    • text :LLM 回答内容。当 LLM 输出为流式时,此处为 100 字为一个分段的内容。

    • ModerationOutputsResult

      • flagged 是否违反校验规则

      • action 执行动作

        • direct_output直接输出预设回复

        • overridden覆写传入变量值

      • preset_response 预设回复(仅当 action=direct_output 返回)

      • text 覆写的 LLM 回答内容(仅当 action=overridden 返回)。

\

Last updated