nat.plugins.config_optimizer.prompts.base#

ABC for prompt optimizers.

Classes#

BasePromptOptimizer

Interface that all prompt optimization strategies must implement.

Module Contents#

class BasePromptOptimizer#

Bases: abc.ABC

Interface that all prompt optimization strategies must implement.

Prompt optimizers run after parameter optimization (when both are enabled). The runtime passes base_cfg as the already-tuned config from the numeric phase, plus optional trial_number_offset and frozen_params.

Unlike BaseParameterOptimizer, this interface returns None. Implementations persist the best prompts to disk (e.g. optimized_prompts.json) rather than updating the config in memory. The config is used as input for evaluation but is not mutated.

abstractmethod run( *, 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, ) → None#

Async:

Run prompt optimization. Persists best prompts to disk; returns None.