nat.parameter_optimization.parameter_optimizer

Attributes

logger

Optional eval runtime class.

Functions

_on_numeric_trial_end(→ None)	Build a TrialResult from one numeric-optimisation trial and fire on_trial_end.
_on_numeric_study_end(→ None)	Fire on_study_end for a completed numeric optimisation study.
optimize_parameters(...)	Tune all non-prompt hyper-parameters and persist the best config.

Module Contents

logger

Optional eval runtime class.

_on_numeric_trial_end(

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

trial: Any,

eval_metrics: list[str],

avg_scores: list[float],

suggestions: dict[str, Any],

last_eval_output: Any,

all_scores: list[list[float]],

) → None

Build a TrialResult from one numeric-optimisation trial and fire on_trial_end.

_on_numeric_study_end(

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

best_trial_obj: Any,

eval_metrics: list[str],

n_trials: int,

) → None

Fire on_study_end for a completed numeric optimisation study.

optimize_parameters(

*,

base_cfg: nat.data_models.config.Config,

full_space: collections.abc.Mapping[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,

) → tuple[nat.data_models.config.Config, dict[str, Any], int]

Tune all non-prompt hyper-parameters and persist the best config.