pyrit.prompt_normalizer.PromptNormalizer

pyrit.prompt_normalizer.PromptNormalizer#

class PromptNormalizer(start_token: str = '⟪', end_token: str = '⟫')[source]#

Bases: object

__init__(start_token: str = '⟪', end_token: str = '⟫') → None[source]#

Initializes the PromptNormalizer.

start_token and end_token are used to delineate which part of a prompt is converted.

Methods

__init__([start_token, end_token])

Initializes the PromptNormalizer.

add_prepended_conversation_to_memory(...[, ...])

Processes the prepended conversation by converting it if needed and adding it to memory.

convert_values(converter_configurations, message)

send_prompt_async(*, seed_group, target[, ...])

Sends a single request to a target.

send_prompt_batch_to_target_async(*, ...[, ...])

Sends a batch of prompts to the target asynchronously.

set_skip_criteria(skip_criteria, skip_value_type)

Sets the skip criteria for the attack.
async add_prepended_conversation_to_memory(conversation_id: str, should_convert: bool = True, converter_configurations: list[PromptConverterConfiguration] | None = None, attack_identifier: dict[str, str] | None = None, prepended_conversation: list[Message] | None = None) → list[Message] | None[source]#

Processes the prepended conversation by converting it if needed and adding it to memory.

Parameters:

  • conversation_id (str) – The conversation ID to use for the message pieces

  • should_convert (bool) – Whether to convert the prepended conversation

  • converter_configurations (Optional[list[PromptConverterConfiguration]]) – Configurations for converting the request

  • attack_identifier (Optional[dict[str, str]]) – Identifier for the attack

  • prepended_conversation (Optional[list[Message]]) – The conversation to prepend

Returns:

The processed prepended conversation

Return type:

Optional[list[Message]]

async convert_values(converter_configurations: list[PromptConverterConfiguration], message: Message)[source]#
async send_prompt_async(*, seed_group: SeedGroup, target: PromptTarget, conversation_id: str | None = None, request_converter_configurations: list[PromptConverterConfiguration] = [], response_converter_configurations: list[PromptConverterConfiguration] = [], labels: dict[str, str] | None = None, attack_identifier: dict[str, str] | None = None) → Message[source]#

Sends a single request to a target.

Parameters:

  • seed_group (SeedGroup) – The seed group to be sent.

  • target (PromptTarget) – The target to which the prompt is sent.

  • conversation_id (str, optional) – The ID of the conversation. Defaults to None.

  • request_converter_configurations (list[PromptConverterConfiguration], optional) – Configurations for converting the request. Defaults to an empty list.

  • response_converter_configurations (list[PromptConverterConfiguration], optional) – Configurations for converting the response. Defaults to an empty list.

  • labels (Optional[dict[str, str]], optional) – Labels associated with the request. Defaults to None.

  • attack_identifier (Optional[dict[str, str]], optional) – Identifier for the attack. Defaults to None.

Raises:

  • Exception – If an error occurs during the request processing.

  • ValueError – If the prompts in the SeedGroup are not part of the same sequence.

Returns:

The response received from the target.

Return type:

Message

async send_prompt_batch_to_target_async(*, requests: list[NormalizerRequest], target: PromptTarget, labels: dict[str, str] | None = None, attack_identifier: dict[str, str] | None = None, batch_size: int = 10) → list[Message][source]#

Sends a batch of prompts to the target asynchronously.

Parameters:

  • requests (list[NormalizerRequest]) – A list of NormalizerRequest objects to be sent.

  • target (PromptTarget) – The target to which the prompts are sent.

  • labels (Optional[dict[str, str]], optional) – A dictionary of labels to be included with the request. Defaults to None.

  • attack_identifier (Optional[dict[str, str]], optional) – A dictionary identifying the attack. Defaults to None.

  • batch_size (int, optional) – The number of prompts to include in each batch. Defaults to 10.

Returns:

A list of Message objects representing the responses

received for each prompt.

Return type:

list[Message]

set_skip_criteria(skip_criteria: PromptFilterCriteria, skip_value_type: Literal['original', 'converted'], ensure_response=True) → None[source]#

Sets the skip criteria for the attack.

If prompts match this in memory and are the same as one being sent, then they won’t be sent to a target.

Prompts are the same if either the original prompt or the converted prompt, determined by skip_value_type flag.

Contents