跳至内容

答案相关性

评估模块。

答案相关性评估器 #

基类: BaseEvaluator

答案相关性评估器。

评估响应与查询的相关性。 该评估器会考虑查询字符串和响应字符串。

参数:

名称 类型 描述 默认值
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]]

用于优化的模板。

 required
Source code in llama-index-core/llama_index/core/evaluation/answer_relevancy.py 
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
class AnswerRelevancyEvaluator(BaseEvaluator):
    """
    Answer relevancy evaluator.

    Evaluates the relevancy of response to a query.
    This evaluator considers the query string and response string.

    Args:
        raise_error(Optional[bool]):
            Whether to raise an error if the response is invalid.
            Defaults to False.
        eval_template(Optional[Union[str, BasePromptTemplate]]):
            The template to use for evaluation.
        refine_template(Optional[Union[str, BasePromptTemplate]]):
            The template to use for refinement.

    """

    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,
    ) -> None:
        """Init params."""
        self._llm = llm or Settings.llm
        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:
        """Get prompts."""
        return {
            "eval_template": self._eval_template,
            "refine_template": self._refine_template,
        }

    def _update_prompts(self, prompts: PromptDictType) -> None:
        """Update prompts."""
        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:
        """Evaluate whether the response is relevant to the query."""
        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/llama_index/core/evaluation/answer_relevancy.py 
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
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:
    """Evaluate whether the response is relevant to the query."""
    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,
    )