nat.parameter_optimization.prompt_optimizer

Attributes

logger

Classes

PromptOptimizerInputSchema

Functions

_on_prompt_trial_end(→ None)	Build TrialResults for each individual in a GA generation and fire on_trial_end.
_on_prompt_study_end(→ None)	Fire on_study_end for a completed prompt GA optimisation study.
optimize_prompts(→ None)

Module Contents

logger

_on_prompt_trial_end(

callback_manager: nat.profiler.parameter_optimization.optimizer_callbacks.OptimizerCallbackManager | None,

population: collections.abc.Sequence[Any],

eval_metrics: list[str],

frozen_params: dict[str, Any] | None,

prompt_format_map: dict[str, str | None],

best: Any,

) → None

Build TrialResults for each individual in a GA generation and fire on_trial_end.

_on_prompt_study_end(

callback_manager: nat.profiler.parameter_optimization.optimizer_callbacks.OptimizerCallbackManager | None,

best: Any,

frozen_params: dict[str, Any] | None,

prompt_format_map: dict[str, str | None],

trial_number_offset: int,

generations: int,

pop_size: int,

) → None

Fire on_study_end for a completed prompt GA optimisation study.

class PromptOptimizerInputSchema(/, **data: Any)

Bases: pydantic.BaseModel

original_prompt: str

objective: str

oracle_feedback: str | None = None

async optimize_prompts(

*,

base_cfg: nat.data_models.config.Config,

full_space: dict[str, nat.data_models.optimizable.SearchSpace],

optimizer_config: nat.data_models.optimizer.OptimizerConfig,

opt_run_config: nat.data_models.optimizer.OptimizerRunConfig,

callback_manager: nat.profiler.parameter_optimization.optimizer_callbacks.OptimizerCallbackManager | None = None,

trial_number_offset: int = 0,

frozen_params: dict[str, Any] | None = None,

) → None