pyrit.executor.promptgen.AnecdoctorGenerator

pyrit.executor.promptgen.AnecdoctorGenerator#

class AnecdoctorGenerator(*, objective_target: PromptChatTarget, processing_model: PromptChatTarget | None = None, converter_config: StrategyConverterConfig | None = None, prompt_normalizer: PromptNormalizer | None = None)[source]#

Bases: PromptGeneratorStrategy[AnecdoctorContext, AnecdoctorResult]

Implementation of the Anecdoctor prompt generation strategy.

The Anecdoctor generator creates misinformation content by either: 1. Using few-shot examples directly (default mode when processing_model is not provided) 2. First extracting a knowledge graph from examples, then using it (when processing_model is provided)

This generator is designed to test a model’s susceptibility to generating false or misleading content when provided with examples in ClaimsReview format. The generator can optionally use a processing model to extract a knowledge graph representation of the examples before generating the final content.

The generation flow consists of: 1. (Optional) Extract knowledge graph from evaluation data using processing model 2. Format a system prompt based on language and content type 3. Send formatted examples (or knowledge graph) to target model 4. Return the generated content as the result

__init__(*, objective_target: PromptChatTarget, processing_model: PromptChatTarget | None = None, converter_config: StrategyConverterConfig | None = None, prompt_normalizer: PromptNormalizer | None = None) → None[source]#

Initialize the Anecdoctor prompt generation strategy.

Parameters:

  • objective_target (PromptChatTarget) – The chat model to be used for prompt generation.

  • processing_model (Optional[PromptChatTarget]) – The model used for knowledge graph extraction. If provided, the generator will extract a knowledge graph from the examples before generation. If None, the generator will use few-shot examples directly.

  • converter_config (Optional[StrategyConverterConfig]) – Configuration for prompt converters.

  • prompt_normalizer (Optional[PromptNormalizer]) – Normalizer for handling prompts.

Methods

__init__(*, objective_target[, ...])

Initialize the Anecdoctor prompt generation strategy.

execute_async()

Execute the prompt generation strategy asynchronously with the provided parameters.

execute_with_context_async(*, context)

Execute strategy with complete lifecycle management.

get_identifier()

Get a serializable identifier for the strategy instance.
async execute_async(*, content_type: str, language: str, evaluation_data: List[str], memory_labels: dict[str, str] | None = None, **kwargs) → AnecdoctorResult[source]#
async execute_async(**kwargs) → AnecdoctorResult

Execute the prompt generation strategy asynchronously with the provided parameters.

Parameters:

  • content_type (str) – The type of content to generate (e.g., “viral tweet”, “news article”).

  • language (str) – The language of the content to generate (e.g., “english”, “spanish”).

  • evaluation_data (List[str]) – The data in ClaimsReview format to use in constructing the prompt.

  • memory_labels (Optional[Dict[str, str]]) – Memory labels for the generation context.

  • **kwargs – Additional parameters for the generation.

Returns:

The result of the anecdoctor generation.

Return type:

AnecdoctorResult

Contents