autointent.generation.utterances.UtteranceEvolver#

class autointent.generation.utterances.UtteranceEvolver(generator, prompt_makers, seed=0, async_mode=False)#

Evolutionary strategy to augmenting utterances.

Deeply inspired by DeepEval evolutions. This method takes single utterance and prompts LLM to change it in a specific way.

Parameters:

generator (autointent.generation.Generator) – Generator instance for generating utterances.
prompt_makers (collections.abc.Sequence[autointent.generation.chat_templates.EvolutionChatTemplate]) – List of prompt makers for generating prompts.
seed (int) – Random seed for reproducibility.
async_mode (bool) – Whether to use asynchronous mode for generation.

Usage#

from autointent import Dataset
from autointent.generation import Generator
from autointent.generation.utterances import UtteranceEvolver
from autointent.generation.chat_templates import GoofyEvolution, InformalEvolution

dataset = Dataset.from_json(path_to_json)
generator = Generator()
evolver = UtteranceEvolver(generator, prompt_makers=[GoofyEvolution(), InformalEvolution()])
evolver.augment(dataset)

generator#

prompt_makers#

async_mode = False#

__call__(utterance, intent_data, n_evolutions=1, sequential=False)#

Apply evolutions to the utterance.

Parameters:

utterance (str) – Utterance to be evolved.
intent_data (autointent.schemas.Intent) – Intent data for which to evolve the utterance.
n_evolutions (int) – Number of evolutions to apply.
sequential (bool) – Whether to apply evolutions sequentially.

Return type:

list[str]

augment(dataset, split_name=Split.TRAIN, n_evolutions=1, update_split=True, batch_size=4, sequential=False)#

Add LLM-generated samples to some split of dataset.

Parameters:

dataset (autointent.Dataset) – Dataset object.
split_name (str) – Dataset split (default is TRAIN).
n_evolutions (int) – Number of evolutions to apply.
update_split (bool) – Whether to update the dataset split.
batch_size (int) – Batch size for async generation.
sequential (bool) – Whether to apply evolutions sequentially.

Return type:

datasets.Dataset