Generative AI Moderation

class datarobot.models.moderation.configuration.ModerationConfiguration

Details of overall moderation configuration for a model.

classmethod get(config_id)

Get a guard configuration by ID.

Added in version v3.6.

Parameters:

config_id (str) – ID of the configuration.

Returns:

Retrieved configuration.

Return type:

ModerationConfiguration

Raises:
classmethod list(entity_id, entity_type)

Returns a list of guard configurations.

Added in version v3.6.

Parameters:

  • entity_id (str) – ID of the entity.

  • entity_type (ModerationGuardEntityType) – Type of the entity.

Returns:

a list of configurations

Return type:

List[ModerationConfiguration]

Raises:
classmethod create(template_id, name, description, stages, entity_id, entity_type, intervention=None, llm_type=None, deployment_id=None, model_info=None, nemo_info=None, openai_api_key='', openai_api_base='', openai_deployment_id='', openai_credential='')

Create a configuration. This is not a full create from scratch; it’s based on a template.

Added in version v3.6.

Parameters:

  • template_id (str) – ID of the template to base this configuration on.

  • name (str) – Name of the configuration.

  • description (str) – Description of the configuration.

  • stages (List[ModerationGuardStage]) – The stages of moderation where this guard is active.

  • entity_id (str) – ID of the custom model version or playground this configuration applies to.

  • entity_type (ModerationGuardEntityType) – Type of the associated entity_id.

  • llm_type (Optional[ModerationGuardLlmType]) – The backing LLM this guard uses.

  • nemo_info (Optional[GuardNemoInfo]) – Additional configuration for NeMo Guardrails guards.

  • model_info (Optional[GuardModelInfo]) – Additional configuration for guards using a deployed model.

  • intervention (Optional[GuardInterventionForConfiguration]) – The assessment conditions, and action the guard should take if conditions are met.

  • openai_api_key (str) – Token to use for OpenAI. Deprecated; use openai_credential instead.

  • openai_api_base (Optional[str]) – Base of the OpenAI connection.

  • openai_deployment_id (Optional[str]) – ID of the OpenAI deployment.

  • openai_credential (Optional[str]) – ID of the credential defined in DataRobot for OpenAI.

Returns:

Created ModerationConfiguration.

Return type:

ModerationConfiguration

Raises:
update(name=None, description=None, intervention=None, llm_type=None, deployment_id=None, model_info=None, nemo_info=None, openai_api_key='', openai_api_base='', openai_deployment_id='', openai_credential='')

Updates the configuration. All fields are optional, and omitted fields are left unchanged.

entity_id, entity_type, and stages cannot be modified for a guard configuration,.

Added in version v3.6.

Parameters:

  • name (str) – Name of the configuration.

  • description (str) – Description of the configuration.

  • llm_type (Optional[ModerationGuardLlmType]) – The backing LLM this guard uses.

  • nemo_info (Optional[GuardNemoInfo]) – Additional configuration for NeMo Guardrails guards.

  • model_info (Optional[GuardModelInfo]) – Additional configuration for guards using a deployed model.

  • intervention (Optional[GuardInterventionForConfiguration]) – The assessment conditions, and action the guard should take if conditions are met.

  • openai_api_key (str) – Token to use for OpenAI. Deprecated; use openai_credential instead.

  • openai_api_base (Optional[str]) – Base of the OpenAI connection.

  • openai_deployment_id (Optional[str]) – ID of the OpenAI deployment.

  • openai_credential (Optional[str]) – ID of the credential defined in DataRobot for OpenAI.

Raises:
Return type:

None

refresh()

Updates the moderation configuration with the latest data from server.

Added in version v3.6.

Raises:
Return type:

None

delete()

Deletes the configuration.

Added in version v3.6.

Raises:
Return type:

None

class datarobot.models.moderation.intervention.GuardInterventionCondition

Defines a condition for intervention.

class datarobot.models.moderation.intervention.GuardInterventionForTemplate

Defines the intervention conditions and actions a guard can take. Configuration schema differs slightly from template because changes were requested after templates were baked in.

classmethod ensure_object(maybe_dict)

Intervention may arrive as an object, or as a dict. Return an object.

Return type:

GuardInterventionForTemplate

class datarobot.models.moderation.intervention.GuardInterventionForConfiguration

Defines the intervention conditions and actions a guard can take. Configuration schema differs slightly from template because changes were requested after templates were baked in.

classmethod ensure_object(maybe_dict)

Intervention may arrive as an object, or as a dict. Return an object.

Return type:

GuardInterventionForConfiguration

class datarobot.models.moderation.model_info.GuardModelInfo

Model information for moderation templates and configurations. Omitted optional values are stored and presented as: * [] (for class names) * None (all others)

to_dict()

Convert the model information object to a dictionary.

Return type:

Dict[str, Union[str, List[str], None]]

classmethod ensure_object(maybe_dict)

Model info may arrive as an object, or as a dict. Return an object.

Return type:

GuardModelInfo

class datarobot.models.moderation.model_version_update.ModelVersionUpdate

Implements the save-configuration operation from the moderation UI. All guard configurations and overall config is saved to a new custom model version.

classmethod new_custom_model_version_with_config(custom_model_id, overall_config, configs)

Create a new custom model version with the provided moderation configuration :type custom_model_id: str :param custom_model_id: :type overall_config: OverallModerationConfig :param overall_config: :type configs: List[ModerationConfiguration] :param configs:

Return type:

ID of the new custom model version.

class datarobot.models.moderation.nemo_info.GuardNemoInfo

Details of a NeMo Guardrails moderation guard.

class datarobot.models.moderation.overall.OverallModerationConfig

Details of overall moderation configuration for a model.

classmethod find(entity_id, entity_type)

Find overall configuration by entity ID and entity type. Each entity (such as a customModelVersion) may have at most 1 overall moderation configuration.

Added in version v3.6.

Parameters:

  • entity_id (str) – ID of the entity.

  • entity_type (str) – Type of the entity.

Returns:

an OverallModerationConfig or None

Return type:

Optional[OverallModerationConfig]

Raises:
classmethod locate(entity_id, entity_type)

Find overall configuration by entity ID and entity type. This version of find() expects the object to exist. Its return type is not optional.

Added in version v3.6.

Parameters:

  • entity_id (str) – ID of the entity.

  • entity_type (str) – Type of the entity.

Returns:

a list of OverallModerationConfig

Return type:

List[OverallModerationConfig]

Raises:
classmethod create(timeout_sec, timeout_action, entity_id, entity_type)

Create an OverallModerationConfig.

Added in version v3.6.

Parameters:

  • timeout_sec (int) – How long to wait for all moderation tasks in a phase to complete.

  • timeout_action (ModerationTimeoutActionType) – What to do if moderation times out.

  • entity_id (str) – Entity, such as customModelVersion, that this configuration applies to.

  • entity_type (ModerationGuardEntityType) – Type of the entity defined by entity_id.

Returns:

created OverallModerationConfig

Return type:

OverallModerationConfig

Raises:
update(timeout_sec, timeout_action, entity_id, entity_type)

Update an OverallModerationConfig.

Added in version v3.6.

Parameters:

  • timeout_sec (int) – How long to wait for all moderation tasks in a phase to complete.

  • timeout_action (ModerationTimeoutActionType) – What to do if moderation times out.

  • entity_id (str) – Entity, such as customModelVersion, that this configuration applies to.

  • entity_type (ModerationGuardEntityType) – Type of the entity defined by entity_id.

Returns:

created OverallModerationConfig

Return type:

OverallModerationConfig

Raises:
refresh()

Updates the overall moderation configuration with the latest data from server.

Added in version v3.6.

Raises:
Return type:

None

class datarobot.models.moderation.template.ModerationTemplate

A DataRobot Moderation Template.

Added in version v3.6.

Variables:

  • id (str) – The ID of the template.

  • name (str) – The name of the template.

classmethod get(template_id)

Returns the template by ID.

Added in version v3.6.

Parameters:

template_id (str) – The ID of the template.

Returns:

The retrieved template.

Return type:

ModerationTemplate

Raises:
classmethod list(include_agentic=None, is_agentic=None, for_playground=None, for_production=None, name=None)

Returns a list of templates.

Added in version v3.6.

Parameters:

yet (none)

Returns:

A list of templates.

Return type:

List[ModerationTemplate]

Raises:
classmethod find(name)

Finds the template by name.

Added in version v3.6.

Parameters:

name (str) – The name of the template.

Returns:

The matching template, or None if no template is found.

Return type:

ModerationTemplate

Raises:
classmethod create(name, description, type, allowed_stages, intervention=None, ootb_type=None, llm_type=None, model_info=None, nemo_info=None)

Creates a template.

Added in version v3.6.

Parameters:

  • name (str) – The name of the template.

  • description (str) – The description of the template.

  • type (GuardType) – The type of the template.

  • allowed_stages (List[ModerationGuardStage]) – The stages of moderation where this guard may be used.

  • ootb_type (ModerationGuardOotbType) – For guards of type ootb, the specific Out of the Box metric type.

  • llm_type (Optional[ModerationGuardLlmType]) – The backing LLM this guard uses.

  • nemo_info (Optional[GuardNemoInfo]) – Additional configuration for NeMo Guardrails guards.

  • model_info (Optional[GuardModelInfo]) – Additional configuration for guards using a deployed model.

  • intervention (Optional[GuardInterventionForTemplate]) – The assessment conditions and action the guard should take if conditions are met.

Returns:

The created template.

Return type:

ModerationTemplate

Raises:
update(name=None, description=None, type=None, allowed_stages=None, intervention=None, ootb_type=None, llm_type=None, model_info=None, nemo_info=None)

Updates the template. All fields are optional, and omitted fields are left unchanged.

Added in version v3.6.

Parameters:

  • name (str) – The name of the template.

  • description (str) – The description of the template.

  • type (GuardType) – The type of the template.

  • allowed_stages (List[ModerationGuardStage]) – The stages of moderation where this guard may be used.

  • ootb_type (ModerationGuardOotbType) – For guards of type ootb, the specific Out of the Box metric type.

  • llm_type (Optional[ModerationGuardLlmType]) – The backing LLM this guard uses.

  • nemo_info (Optional[GuardNemoInfo]) – Additional configuration for NeMo Guardrails guards.

  • model_info (Optional[GuardModelInfo]) – Additional configuration for guards using a deployed model.

  • intervention (Optional[GuardInterventionForTemplate]) – The assessment conditions and action the guard should take if conditions are met.

Raises:
Return type:

None

refresh()

Updates the template with the latest data from the server.

Added in version v3.6.

Raises:
Return type:

None

delete()

Deletes the template.

Added in version v3.6.

Raises:
Return type:

None

class datarobot.enums.ModerationGuardOotbType

Defines the available OOTB guards.