pyrit.models.SeedDataset

class SeedDataset(*, seeds: Sequence[Dict[str, Any]] | Sequence[Seed] | None = None, data_type: Literal['text', 'image_path', 'audio_path', 'video_path', 'binary_path', 'url', 'reasoning', 'error', 'function_call', 'tool_call', 'function_call_output'] | None = 'text', name: str | None = None, dataset_name: str | None = None, harm_categories: Sequence[str] | None = None, description: str | None = None, authors: Sequence[str] | None = None, groups: Sequence[str] | None = None, source: str | None = None, date_added: datetime | None = None, added_by: str | None = None, seed_type: Literal['prompt', 'objective', 'simulated_conversation'] | None = None, is_objective: bool = False)

Bases: YamlLoadable

SeedDataset manages seed prompts plus optional top-level defaults. Prompts are stored as a Sequence[Seed], so references to prompt properties are straightforward (e.g. ds.seeds[0].value).

__init__(*, seeds: Sequence[Dict[str, Any]] | Sequence[Seed] | None = None, data_type: Literal['text', 'image_path', 'audio_path', 'video_path', 'binary_path', 'url', 'reasoning', 'error', 'function_call', 'tool_call', 'function_call_output'] | None = 'text', name: str | None = None, dataset_name: str | None = None, harm_categories: Sequence[str] | None = None, description: str | None = None, authors: Sequence[str] | None = None, groups: Sequence[str] | None = None, source: str | None = None, date_added: datetime | None = None, added_by: str | None = None, seed_type: Literal['prompt', 'objective', 'simulated_conversation'] | None = None, is_objective: bool = False)

Initialize the dataset. Typically, you’ll call from_dict or from_yaml_file so that top-level defaults are merged into each seed. If you’re passing seeds directly, they can be either a list of Seed objects or seed dictionaries (which then get converted to Seed objects).

Parameters:

• seeds – List of seed dictionaries or Seed objects.

• data_type – Default data type for seeds.

• name – Name of the dataset.

• dataset_name – Dataset name for categorization.

• harm_categories – List of harm categories.

• description – Description of the dataset.

• authors – List of authors.

• groups – List of groups.

• source – Source of the dataset.

• date_added – Date when the dataset was added.

• added_by – User who added the dataset.

• seed_type – The type of seeds in this dataset (“prompt”, “objective”, or “simulated_conversation”).

• is_objective – Deprecated in 0.13.0. Use seed_type=”objective” instead.

Methods

__init__(*[, seeds, data_type, name, ...])	Initialize the dataset.
from_dict(data)	Builds a SeedDataset by merging top-level defaults into each item in 'seeds'.
from_yaml_file(file)	Create a new object from a YAML file.
get_random_values(*, number[, harm_categories])	Extracts and returns a list of random prompt values from the dataset.
get_values(*[, first, last, harm_categories])	Extracts and returns a list of prompt values from the dataset.
group_seed_prompts_by_prompt_group_id(seeds)	Groups the given list of Seeds by their prompt_group_id and creates SeedGroup or SeedAttackGroup instances.
render_template_value(**kwargs)	Renders self.value as a template, applying provided parameters in kwargs.

Attributes

objectives
prompts
seed_groups	Returns the seeds grouped by their prompt_group_id.
data_type
name
dataset_name
harm_categories
description
authors
groups
source
date_added
added_by
seeds

added_by: str | None

authors: Sequence[str] | None

data_type: str | None

dataset_name: str | None

date_added: datetime | None

description: str | None

classmethod from_dict(data: Dict[str, Any]) → SeedDataset

Builds a SeedDataset by merging top-level defaults into each item in ‘seeds’.

get_random_values(*, number: Annotated[int, Gt(gt=0)], harm_categories: Sequence[str] | None = None) → Sequence[str]

Extracts and returns a list of random prompt values from the dataset.

Parameters:

• number (int) – The number of random prompt values to return.

• harm_categories (Optional[Sequence[str]]) – If provided, only prompts containing at least one of these harm categories are included.

Returns:

A list of prompt values.

Return type:

Sequence[str]

get_values(*, first: Annotated[int, Gt(gt=0)] | None = None, last: Annotated[int, Gt(gt=0)] | None = None, harm_categories: Sequence[str] | None = None) → Sequence[str]

Extracts and returns a list of prompt values from the dataset. By default, returns all of them.

Parameters:

• first (Optional[int]) – If provided, values from the first N prompts are included.

• last (Optional[int]) – If provided, values from the last N prompts are included.

• harm_categories (Optional[Sequence[str]]) – If provided, only prompts containing at least one of these harm categories are included.

Returns:

A list of prompt values.

Return type:

Sequence[str]

static group_seed_prompts_by_prompt_group_id(seeds: Sequence[Seed]) → Sequence[SeedGroup]

Groups the given list of Seeds by their prompt_group_id and creates SeedGroup or SeedAttackGroup instances.

For each group, this method first attempts to create a SeedAttackGroup (which has attack-specific properties like objective). If validation fails, it falls back to a basic SeedGroup.

Parameters:

seeds – A list of Seed objects.

Returns:

A list of SeedGroup or SeedAttackGroup objects, with seeds grouped by prompt_group_id. Each group will be ordered by the sequence number of the seeds, if available.

groups: Sequence[str] | None

harm_categories: Sequence[str] | None

name: str | None

property objectives: Sequence[SeedObjective]

property prompts: Sequence[SeedPrompt]

render_template_value(**kwargs: object) → None

Renders self.value as a template, applying provided parameters in kwargs.

Parameters:

kwargs – Key-value pairs to replace in the SeedDataset value.

Returns:

None

Raises:

ValueError – If parameters are missing or invalid in the template.

property seed_groups: Sequence[SeedGroup]

Returns the seeds grouped by their prompt_group_id.

Returns:

A list of SeedGroup objects, with seeds grouped by prompt_group_id.

Return type:

Sequence[SeedGroup]

seeds: Sequence[Seed]

source: str | None