pyrit.executor.attack.PrependedConversationConfig

pyrit.executor.attack.PrependedConversationConfig#

class PrependedConversationConfig(apply_converters_to_roles: ~typing.List[~typing.Literal['system', 'user', 'assistant', 'simulated_assistant', 'tool', 'developer']] = <factory>, message_normalizer: ~pyrit.message_normalizer.message_normalizer.MessageStringNormalizer | None = None, non_chat_target_behavior: ~typing.Literal['normalize_first_turn', 'raise'] = 'normalize_first_turn')[source]#

Bases: object

Configuration for controlling how prepended conversations are processed before being sent to the objective target.

This class provides control over: - Which message roles should have request converters applied - How to normalize conversation history for non-chat objective targets - What to do when the objective target is not a PromptChatTarget

__init__(apply_converters_to_roles: ~typing.List[~typing.Literal['system', 'user', 'assistant', 'simulated_assistant', 'tool', 'developer']] = <factory>, message_normalizer: ~pyrit.message_normalizer.message_normalizer.MessageStringNormalizer | None = None, non_chat_target_behavior: ~typing.Literal['normalize_first_turn', 'raise'] = 'normalize_first_turn') → None#

Methods

__init__([apply_converters_to_roles, ...])

default()

Create a default configuration with converters applied to all roles.

for_non_chat_target(*[, message_normalizer, ...])

Create a configuration for use with non-chat targets.

get_message_normalizer()

Get the normalizer for objective target context, with a default fallback.

Attributes

message_normalizer

non_chat_target_behavior

apply_converters_to_roles

apply_converters_to_roles: List[Literal['system', 'user', 'assistant', 'simulated_assistant', 'tool', 'developer']]#
classmethod default() → PrependedConversationConfig[source]#

Create a default configuration with converters applied to all roles.

Returns:

A configuration that applies converters to all prepended messages, raising an error for non-chat targets.

classmethod for_non_chat_target(*, message_normalizer: MessageStringNormalizer | None = None, apply_converters_to_roles: List[Literal['system', 'user', 'assistant', 'simulated_assistant', 'tool', 'developer']] | None = None) → PrependedConversationConfig[source]#

Create a configuration for use with non-chat targets.

This configuration normalizes the prepended conversation into a text block that will be prepended to the first message sent to the target.

Parameters:

  • message_normalizer – Normalizer for formatting the prepended conversation into a string. Defaults to ConversationContextNormalizer if not provided.

  • apply_converters_to_roles – Roles to apply converters to before normalization. Defaults to all roles.

Returns:

A configuration that normalizes the prepended conversation for non-chat targets.

get_message_normalizer() → MessageStringNormalizer[source]#

Get the normalizer for objective target context, with a default fallback.

Returns:

The configured objective_target_context_normalizer, or a default ConversationContextNormalizer if none was configured.

message_normalizer: MessageStringNormalizer | None = None#
non_chat_target_behavior: Literal['normalize_first_turn', 'raise'] = 'normalize_first_turn'#
Contents