dspy.ts/dist/optimize/base.d.ts

DSPy.ts - Declarative Self-Learning TypeScript: A framework for compositional LM pipelines with self-improving prompt strategies.

/**
 * Base classes and types for DSPy.ts optimizers
 */
import { Module } from '../core/module';
import { Pipeline } from '../core/pipeline';
/**
 * Metric function type for evaluating program outputs
 */
export type MetricFunction<TInput = any, TOutput = any> = (input: TInput, output: TOutput, expected?: TOutput) => number;
/**
 * Base optimizer configuration
 */
export interface OptimizerConfig {
    maxIterations?: number;
    numThreads?: number;
    debug?: boolean;
}
/**
 * Training example type
 */
export interface TrainingExample<TInput = any, TOutput = any> {
    input: TInput;
    output?: TOutput;
}
/**
 * Base class for all DSPy.ts optimizers
 */
export declare abstract class Optimizer<TInput = any, TOutput = any> {
    protected config: Required<OptimizerConfig>;
    protected metric: MetricFunction<TInput, TOutput>;
    constructor(metric: MetricFunction<TInput, TOutput>, config?: OptimizerConfig);
    /**
     * Compile a program or module with optimization
     */
    abstract compile(program: Module<any, any> | Pipeline, trainset: TrainingExample<TInput, TOutput>[]): Promise<Module<any, any> | Pipeline>;
    /**
     * Save the optimized program to a file
     */
    abstract save(path: string, saveFieldMeta?: boolean): void;
    /**
     * Load an optimized program from a file
     */
    abstract load(path: string): void;
    protected log(message: string): void;
}