feat: add api-based extension & external data tool & moderation backend (#1403)

Co-authored-by: takatost <takatost@gmail.com>

api/core/moderation/api/__builtin__

@@ -0,0 +1 @@
+3

api/core/moderation/api/api.py

@@ -0,0 +1,88 @@
+from pydantic import BaseModel
+
+from core.moderation.base import Moderation, ModerationInputsResult, ModerationOutputsResult, ModerationAction
+from core.extension.api_based_extension_requestor import APIBasedExtensionRequestor, APIBasedExtensionPoint
+from core.helper.encrypter import decrypt_token
+from extensions.ext_database import db
+from models.api_based_extension import APIBasedExtension
+
+
+class ModerationInputParams(BaseModel):
+    app_id: str = ""
+    inputs: dict = {}
+    query: str = ""
+
+
+class ModerationOutputParams(BaseModel):
+    app_id: str = ""
+    text: str
+
+
+class ApiModeration(Moderation):
+    name: str = "api"
+
+    @classmethod
+    def validate_config(cls, tenant_id: str, config: dict) -> None:
+        """
+        Validate the incoming form config data.
+
+        :param tenant_id: the id of workspace
+        :param config: the form config data
+        :return:
+        """
+        cls._validate_inputs_and_outputs_config(config, False)
+
+        api_based_extension_id = config.get("api_based_extension_id")
+        if not api_based_extension_id:
+            raise ValueError("api_based_extension_id is required")
+
+        extension = cls._get_api_based_extension(tenant_id, api_based_extension_id)
+        if not extension:
+            raise ValueError("API-based Extension not found. Please check it again.")
+
+    def moderation_for_inputs(self, inputs: dict, query: str = "") -> ModerationInputsResult:
+        flagged = False
+        preset_response = ""
+
+        if self.config['inputs_config']['enabled']:
+            params = ModerationInputParams(
+                app_id=self.app_id,
+                inputs=inputs,
+                query=query
+            )
+
+            result = self._get_config_by_requestor(APIBasedExtensionPoint.APP_MODERATION_INPUT, params.dict())
+            return ModerationInputsResult(**result)
+
+        return ModerationInputsResult(flagged=flagged, action=ModerationAction.DIRECT_OUTPUT, preset_response=preset_response)
+
+    def moderation_for_outputs(self, text: str) -> ModerationOutputsResult:
+        flagged = False
+        preset_response = ""
+
+        if self.config['outputs_config']['enabled']:
+            params = ModerationOutputParams(
+                app_id=self.app_id,
+                text=text
+            )
+
+            result = self._get_config_by_requestor(APIBasedExtensionPoint.APP_MODERATION_OUTPUT, params.dict())
+            return ModerationOutputsResult(**result)
+
+        return ModerationOutputsResult(flagged=flagged, action=ModerationAction.DIRECT_OUTPUT, preset_response=preset_response)
+
+    def _get_config_by_requestor(self, extension_point: APIBasedExtensionPoint, params: dict) -> dict:
+        extension = self._get_api_based_extension(self.tenant_id, self.config.get("api_based_extension_id"))
+        requestor = APIBasedExtensionRequestor(extension.api_endpoint, decrypt_token(self.tenant_id, extension.api_key))
+
+        result = requestor.request(extension_point, params)
+        return result
+
+    @staticmethod
+    def _get_api_based_extension(tenant_id: str, api_based_extension_id: str) -> APIBasedExtension:
+        extension = db.session.query(APIBasedExtension).filter(
+            APIBasedExtension.tenant_id == tenant_id,
+            APIBasedExtension.id == api_based_extension_id
+        ).first()
+
+        return extension

api/core/moderation/base.py

@@ -0,0 +1,113 @@
+from abc import ABC, abstractmethod
+from typing import Optional
+from pydantic import BaseModel
+from enum import Enum
+
+from core.extension.extensible import Extensible, ExtensionModule
+
+
+class ModerationAction(Enum):
+    DIRECT_OUTPUT = 'direct_output'
+    OVERRIDED = 'overrided'
+
+
+class ModerationInputsResult(BaseModel):
+    flagged: bool = False
+    action: ModerationAction
+    preset_response: str = ""
+    inputs: dict = {}
+    query: str = ""
+
+
+class ModerationOutputsResult(BaseModel):
+    flagged: bool = False
+    action: ModerationAction
+    preset_response: str = ""
+    text: str = ""
+
+
+class Moderation(Extensible, ABC):
+    """
+    The base class of moderation.
+    """
+    module: ExtensionModule = ExtensionModule.MODERATION
+
+    def __init__(self, app_id: str, tenant_id: str, config: Optional[dict] = None) -> None:
+        super().__init__(tenant_id, config)
+        self.app_id = app_id
+
+    @classmethod
+    @abstractmethod
+    def validate_config(cls, tenant_id: str, config: dict) -> None:
+        """
+        Validate the incoming form config data.
+
+        :param tenant_id: the id of workspace
+        :param config: the form config data
+        :return:
+        """
+        raise NotImplementedError
+
+    @abstractmethod
+    def moderation_for_inputs(self, inputs: dict, query: str = "") -> ModerationInputsResult:
+        """
+        Moderation for inputs.
+        After the user inputs, this method will be called to perform sensitive content review
+        on the user inputs and return the processed results.
+
+        :param inputs: user inputs
+        :param query: query string (required in chat app)
+        :return:
+        """
+        raise NotImplementedError
+
+    @abstractmethod
+    def moderation_for_outputs(self, text: str) -> ModerationOutputsResult:
+        """
+        Moderation for outputs.
+        When LLM outputs content, the front end will pass the output content (may be segmented)
+        to this method for sensitive content review, and the output content will be shielded if the review fails.
+
+        :param text: LLM output content
+        :return:
+        """
+        raise NotImplementedError
+
+    @classmethod
+    def _validate_inputs_and_outputs_config(self, config: dict, is_preset_response_required: bool) -> None:
+        # inputs_config
+        inputs_config = config.get("inputs_config")
+        if not isinstance(inputs_config, dict):
+            raise ValueError("inputs_config must be a dict")
+
+        # outputs_config
+        outputs_config = config.get("outputs_config")
+        if not isinstance(outputs_config, dict):
+            raise ValueError("outputs_config must be a dict")
+
+        inputs_config_enabled = inputs_config.get("enabled")
+        outputs_config_enabled = outputs_config.get("enabled")
+        if not inputs_config_enabled and not outputs_config_enabled:
+            raise ValueError("At least one of inputs_config or outputs_config must be enabled")
+
+        # preset_response
+        if not is_preset_response_required:
+            return
+
+        if inputs_config_enabled:
+            if not inputs_config.get("preset_response"):
+                raise ValueError("inputs_config.preset_response is required")
+
+            if len(inputs_config.get("preset_response")) > 100:
+                raise ValueError("inputs_config.preset_response must be less than 100 characters")
+
+        if outputs_config_enabled:
+            if not outputs_config.get("preset_response"):
+                raise ValueError("outputs_config.preset_response is required")
+
+            if len(outputs_config.get("preset_response")) > 100:
+                raise ValueError("outputs_config.preset_response must be less than 100 characters")
+
+
+class ModerationException(Exception):
+    pass

api/core/moderation/factory.py

@@ -0,0 +1,48 @@
+from core.extension.extensible import ExtensionModule
+from core.moderation.base import Moderation, ModerationInputsResult, ModerationOutputsResult
+from extensions.ext_code_based_extension import code_based_extension
+
+
+class ModerationFactory:
+    __extension_instance: Moderation
+
+    def __init__(self, name: str, app_id: str, tenant_id: str, config: dict) -> None:
+        extension_class = code_based_extension.extension_class(ExtensionModule.MODERATION, name)
+        self.__extension_instance = extension_class(app_id, tenant_id, config)
+
+    @classmethod
+    def validate_config(cls, name: str, tenant_id: str, config: dict) -> None:
+        """
+        Validate the incoming form config data.
+
+        :param name: the name of extension
+        :param tenant_id: the id of workspace
+        :param config: the form config data
+        :return:
+        """
+        code_based_extension.validate_form_schema(ExtensionModule.MODERATION, name, config)
+        extension_class = code_based_extension.extension_class(ExtensionModule.MODERATION, name)
+        extension_class.validate_config(tenant_id, config)
+
+    def moderation_for_inputs(self, inputs: dict, query: str = "") -> ModerationInputsResult:
+        """
+        Moderation for inputs.
+        After the user inputs, this method will be called to perform sensitive content review
+        on the user inputs and return the processed results.
+
+        :param inputs: user inputs
+        :param query: query string (required in chat app)
+        :return:
+        """
+        return self.__extension_instance.moderation_for_inputs(inputs, query)
+
+    def moderation_for_outputs(self, text: str) -> ModerationOutputsResult:
+        """
+        Moderation for outputs.
+        When LLM outputs content, the front end will pass the output content (may be segmented)
+        to this method for sensitive content review, and the output content will be shielded if the review fails.
+
+        :param text: LLM output content
+        :return:
+        """
+        return self.__extension_instance.moderation_for_outputs(text)

api/core/moderation/keywords/__builtin__

@@ -0,0 +1 @@
+2

api/core/moderation/keywords/keywords.py

@@ -0,0 +1,60 @@
+from core.moderation.base import Moderation, ModerationInputsResult, ModerationOutputsResult, ModerationAction
+
+
+class KeywordsModeration(Moderation):
+    name: str = "keywords"
+
+    @classmethod
+    def validate_config(cls, tenant_id: str, config: dict) -> None:
+        """
+        Validate the incoming form config data.
+
+        :param tenant_id: the id of workspace
+        :param config: the form config data
+        :return:
+        """
+        cls._validate_inputs_and_outputs_config(config, True)
+
+        if not config.get("keywords"):
+            raise ValueError("keywords is required")
+
+        if len(config.get("keywords")) > 1000:
+            raise ValueError("keywords length must be less than 1000")
+
+    def moderation_for_inputs(self, inputs: dict, query: str = "") -> ModerationInputsResult:
+        flagged = False
+        preset_response = ""
+
+        if self.config['inputs_config']['enabled']:
+            preset_response = self.config['inputs_config']['preset_response']
+
+            if query:
+                inputs['query__'] = query
+            keywords_list = self.config['keywords'].split('\n')
+            flagged = self._is_violated(inputs, keywords_list)
+
+        return ModerationInputsResult(flagged=flagged, action=ModerationAction.DIRECT_OUTPUT, preset_response=preset_response)
+
+    def moderation_for_outputs(self, text: str) -> ModerationOutputsResult:
+        flagged = False
+        preset_response = ""
+
+        if self.config['outputs_config']['enabled']:
+            keywords_list = self.config['keywords'].split('\n')
+            flagged = self._is_violated({'text': text}, keywords_list)
+            preset_response = self.config['outputs_config']['preset_response']
+
+        return ModerationOutputsResult(flagged=flagged, action=ModerationAction.DIRECT_OUTPUT, preset_response=preset_response)
+
+    def _is_violated(self, inputs: dict, keywords_list: list) -> bool:
+        for value in inputs.values():
+            if self._check_keywords_in_value(keywords_list, value):
+                return True
+
+        return False
+
+    def _check_keywords_in_value(self, keywords_list, value):
+        for keyword in keywords_list:
+            if keyword.lower() in value.lower():
+                return True
+        return False

api/core/moderation/openai_moderation/__builtin__

@@ -0,0 +1 @@
+1

api/core/moderation/openai_moderation/openai_moderation.py

@@ -0,0 +1,46 @@
+from core.moderation.base import Moderation, ModerationInputsResult, ModerationOutputsResult, ModerationAction
+from core.model_providers.model_factory import ModelFactory
+
+
+class OpenAIModeration(Moderation):
+    name: str = "openai_moderation"
+
+    @classmethod
+    def validate_config(cls, tenant_id: str, config: dict) -> None:
+        """
+        Validate the incoming form config data.
+
+        :param tenant_id: the id of workspace
+        :param config: the form config data
+        :return:
+        """
+        cls._validate_inputs_and_outputs_config(config, True)
+
+    def moderation_for_inputs(self, inputs: dict, query: str = "") -> ModerationInputsResult:
+        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.DIRECT_OUTPUT, preset_response=preset_response)
+
+    def moderation_for_outputs(self, text: str) -> ModerationOutputsResult:
+        flagged = False
+        preset_response = ""
+
+        if self.config['outputs_config']['enabled']:
+            flagged = self._is_violated({'text': text})
+            preset_response = self.config['outputs_config']['preset_response']
+
+        return ModerationOutputsResult(flagged=flagged, action=ModerationAction.DIRECT_OUTPUT, preset_response=preset_response)
+
+    def _is_violated(self, inputs: dict):
+        text = '\n'.join(inputs.values())
+        openai_moderation = ModelFactory.get_moderation_model(self.tenant_id, "openai", "moderation")
+        is_not_invalid = openai_moderation.run(text)
+        return not is_not_invalid