Refiner

This module defines the main PromptRefiner class and associated functions for modifying and enhancing prompts using various strategies.

PromptRefiner(strategies)

Bases: object

A class for refining prompt using multiple strategies.

This class applies a series of transformations to enhance a given input prompt. It supports multiple strategies, such as "few_shot" and "meta".

Attributes:

Name	Type	Description
strategies	list[str | BaseStrategy]	A list of strategies to be applied to the prompt.

Initialize the PromptRefiner with list of strategies.

Parameters:

Name	Type	Description	Default
strategies	list	A list of supported strategies.	required

Raises:

Type	Description
ValueError	If provided strategy is not supported.

Source code in promptrefiner/refiner.py

def __init__(self, strategies: list):
    """
    Initialize the `PromptRefiner` with list of strategies.

    Args:
        strategies (list): A list of supported strategies.

    Raises:
        ValueError: If provided strategy is not supported.
    """
    if not strategies:
        raise ValueError("At least one strategy must be provided.")

    self.strategies = []
    for strategy in strategies:
        if isinstance(strategy, str):
            normalized_name = next(
                (
                    name
                    for name, data in STRATEGY_MAP.items()
                    if strategy.lower() in [name] + data["aliases"]
                ),
                None,
            )
            if not normalized_name:
                raise ValueError(f"Unsupported strategy: {strategy}")
            strategy_class = STRATEGY_MAP[normalized_name]["class_"]
            self.strategies.append(strategy_class())
        elif isinstance(strategy, type) and issubclass(strategy, BaseStrategy):
            self.strategies.append(strategy())  # Instantiate class
        elif isinstance(strategy, BaseStrategy):
            self.strategies.append(strategy)  # Already instantiated
        else:
            logger.error(f"Invalid strategy: {strategy}")
            raise ValueError(f"Invalid strategy: {strategy}")
    logger.info(f"PromptRefiner initialized with strategies: {self.strategies}")

refine(prompt)

Applies all strategies in sequence and returns refined prompt.

Parameters:

Name	Type	Description	Default
prompt	str	The input prompt.	required

Returns:

Name	Type	Description
str	str	The refined prompt after applying strategies.

Source code in promptrefiner/refiner.py

def refine(self, prompt: str) -> str:
    """
    Applies all strategies in sequence and returns refined prompt.

    Args:
        prompt (str): The input prompt.

    Returns:
        str: The refined prompt after applying strategies.
    """
    for strategy in self.strategies:
        prompt = strategy.refine(prompt)
    return prompt