latent.optimize.ace.optimizer

ACE-based prompt optimizer.

Classes

ACEOptimizer

ACEOptimizer(agent: BaseAgent, lower_is_better: bool = False, adapter_config: ACEAdapterConfig | None = None, sections: list[SectionConfig] | None = None, tracker: ExperimentTracker | None = None)

Prompt optimizer using the ACE (Automatic Calibration Engine) framework.

Runs N adaptation rounds using the ACE skillbook. Each round: 1. Gets current prompt from skillbook 2. Evaluates agent on train_data using the current prompt 3. Records result via tracker 4. Adapts skillbook based on feedback

Methods

ACEOptimizer.optimize

optimize(train_data: list[dict[str, Any]], metric: Any, max_iterations: int = 10) -> OptimizationResult

Run ACE adaptation rounds.

Args: train_data: List of dicts with input data. metric: A callable (expected_output, predicted_output) -> float. max_iterations: Number of adaptation rounds.

Returns: OptimizationResult with all experiments recorded.