Answer relevancy

评估模块。

AnswerRelevancyEvaluator

Bases: BaseEvaluator

答案相关性评估器。

评估响应对查询的相关性。 该评估器考虑查询字符串和响应字符串。

Parameters:

Name	Type	Description	Default
service_context(Optional[ServiceContext])		用于评估的服务上下文。	required
raise_error(Optional[bool])		如果响应无效是否引发错误。 默认为False。	required
eval_template(Optional[Union[str,	BasePromptTemplate]]	用于评估的模板。	required
refine_template(Optional[Union[str,	BasePromptTemplate]]	用于改进的模板。	required

Source code in llama_index/core/evaluation/answer_relevancy.py

class AnswerRelevancyEvaluator(BaseEvaluator):
    """答案相关性评估器。

    评估响应对查询的相关性。
    该评估器考虑查询字符串和响应字符串。

    Args:
        service_context(Optional[ServiceContext]):
            用于评估的服务上下文。
        raise_error(Optional[bool]):
            如果响应无效是否引发错误。
            默认为False。
        eval_template(Optional[Union[str, BasePromptTemplate]]):
            用于评估的模板。
        refine_template(Optional[Union[str, BasePromptTemplate]]):
            用于改进的模板。"""

    def __init__(
        self,
        llm: Optional[LLM] = None,
        raise_error: bool = False,
        eval_template: str | BasePromptTemplate | None = None,
        score_threshold: float = _DEFAULT_SCORE_THRESHOLD,
        parser_function: Callable[
            [str], Tuple[Optional[float], Optional[str]]
        ] = _default_parser_function,
        # deprecated
        service_context: Optional[ServiceContext] = None,
    ) -> None:
        """初始化参数。"""
        self._llm = llm or llm_from_settings_or_context(Settings, service_context)
        self._raise_error = raise_error

        self._eval_template: BasePromptTemplate
        if isinstance(eval_template, str):
            self._eval_template = PromptTemplate(eval_template)
        else:
            self._eval_template = eval_template or DEFAULT_EVAL_TEMPLATE

        self.parser_function = parser_function
        self.score_threshold = score_threshold

    def _get_prompts(self) -> PromptDictType:
        """获取提示。"""
        return {
            "eval_template": self._eval_template,
            "refine_template": self._refine_template,
        }

    def _update_prompts(self, prompts: PromptDictType) -> None:
        """更新提示。"""
        if "eval_template" in prompts:
            self._eval_template = prompts["eval_template"]
        if "refine_template" in prompts:
            self._refine_template = prompts["refine_template"]

    async def aevaluate(
        self,
        query: str | None = None,
        response: str | None = None,
        contexts: Sequence[str] | None = None,
        sleep_time_in_seconds: int = 0,
        **kwargs: Any,
    ) -> EvaluationResult:
        """评估响应是否与查询相关。"""
        del kwargs  # Unused
        del contexts  # Unused

        if query is None or response is None:
            raise ValueError("query and response must be provided")

        await asyncio.sleep(sleep_time_in_seconds)

        eval_response = await self._llm.apredict(
            prompt=self._eval_template,
            query=query,
            response=response,
        )

        score, reasoning = self.parser_function(eval_response)

        invalid_result, invalid_reason = False, None
        if score is None and reasoning is None:
            if self._raise_error:
                raise ValueError("The response is invalid")
            invalid_result = True
            invalid_reason = "Unable to parse the output string."

        if score:
            score /= self.score_threshold

        return EvaluationResult(
            query=query,
            response=response,
            score=score,
            feedback=eval_response,
            invalid_result=invalid_result,
            invalid_reason=invalid_reason,
        )

aevaluate async

aevaluate(
    query: str | None = None,
    response: str | None = None,
    contexts: Sequence[str] | None = None,
    sleep_time_in_seconds: int = 0,
    **kwargs: Any
) -> EvaluationResult

评估响应是否与查询相关。

Source code in llama_index/core/evaluation/answer_relevancy.py

async def aevaluate(
    self,
    query: str | None = None,
    response: str | None = None,
    contexts: Sequence[str] | None = None,
    sleep_time_in_seconds: int = 0,
    **kwargs: Any,
) -> EvaluationResult:
    """评估响应是否与查询相关。"""
    del kwargs  # Unused
    del contexts  # Unused

    if query is None or response is None:
        raise ValueError("query and response must be provided")

    await asyncio.sleep(sleep_time_in_seconds)

    eval_response = await self._llm.apredict(
        prompt=self._eval_template,
        query=query,
        response=response,
    )

    score, reasoning = self.parser_function(eval_response)

    invalid_result, invalid_reason = False, None
    if score is None and reasoning is None:
        if self._raise_error:
            raise ValueError("The response is invalid")
        invalid_result = True
        invalid_reason = "Unable to parse the output string."

    if score:
        score /= self.score_threshold

    return EvaluationResult(
        query=query,
        response=response,
        score=score,
        feedback=eval_response,
        invalid_result=invalid_result,
        invalid_reason=invalid_reason,
    )