173 lines
6.7 KiB
Python
173 lines
6.7 KiB
Python
"""Relevancy evaluation."""
|
|
from __future__ import annotations
|
|
|
|
import asyncio
|
|
import re
|
|
from typing import Any, Callable, Optional, Sequence, Tuple
|
|
|
|
from llama_index import ServiceContext
|
|
from llama_index.evaluation.base import BaseEvaluator, EvaluationResult
|
|
from llama_index.indices import SummaryIndex
|
|
from llama_index.prompts import BasePromptTemplate, PromptTemplate
|
|
from llama_index.prompts.mixin import PromptDictType
|
|
from llama_index.schema import Document
|
|
|
|
DEFAULT_EVAL_TEMPLATE = PromptTemplate(
|
|
"Your task is to evaluate if the retrieved context from the document sources are relevant to the query.\n"
|
|
"The evaluation should be performed in a step-by-step manner by answering the following questions:\n"
|
|
"1. Does the retrieved context match the subject matter of the user's query?\n"
|
|
"2. Can the retrieved context be used exclusively to provide a full answer to the user's query?\n"
|
|
"Each question above is worth 2 points, where partial marks are allowed and encouraged. Provide detailed feedback on the response "
|
|
"according to the criteria questions previously mentioned. "
|
|
"After your feedback provide a final result by strictly following this format: "
|
|
"'[RESULT] followed by the float number representing the total score assigned to the response'\n\n"
|
|
"Query: \n {query_str}\n"
|
|
"Context: \n {context_str}\n"
|
|
"Feedback:"
|
|
)
|
|
|
|
_DEFAULT_SCORE_THRESHOLD = 4.0
|
|
|
|
DEFAULT_REFINE_TEMPLATE = PromptTemplate(
|
|
"We want to understand if the following query and response is"
|
|
"in line with the context information: \n {query_str}\n"
|
|
"We have provided an existing evaluation score: \n {existing_answer}\n"
|
|
"We have the opportunity to refine the existing evaluation "
|
|
"(only if needed) with some more context below.\n"
|
|
"------------\n"
|
|
"{context_msg}\n"
|
|
"------------\n"
|
|
f"If the existing evaluation was already {_DEFAULT_SCORE_THRESHOLD}, still answer {_DEFAULT_SCORE_THRESHOLD}. "
|
|
f"If the information is present in the new context, answer {_DEFAULT_SCORE_THRESHOLD}. "
|
|
"Otherwise answer {existing_answer}.\n"
|
|
)
|
|
|
|
|
|
def _default_parser_function(output_str: str) -> Tuple[Optional[float], Optional[str]]:
|
|
# Pattern to match the feedback and response
|
|
# This pattern looks for any text ending with '[RESULT]' followed by a number
|
|
pattern = r"([\s\S]+)(?:\[RESULT\]\s*)([\d.]+)"
|
|
|
|
# Using regex to find all matches
|
|
result = re.search(pattern, output_str)
|
|
|
|
# Check if any match is found
|
|
if result:
|
|
# Assuming there's only one match in the text, extract feedback and response
|
|
feedback, score = result.groups()
|
|
score = float(score) if score is not None else score
|
|
return score, feedback.strip()
|
|
else:
|
|
return None, None
|
|
|
|
|
|
class ContextRelevancyEvaluator(BaseEvaluator):
|
|
"""Context relevancy evaluator.
|
|
|
|
Evaluates the relevancy of retrieved contexts to a query.
|
|
This evaluator considers the query string and retrieved contexts.
|
|
|
|
Args:
|
|
service_context(Optional[ServiceContext]):
|
|
The service context to use for evaluation.
|
|
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,
|
|
service_context: ServiceContext | None = None,
|
|
raise_error: bool = False,
|
|
eval_template: str | BasePromptTemplate | None = None,
|
|
refine_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._service_context = service_context or ServiceContext.from_defaults()
|
|
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._refine_template: BasePromptTemplate
|
|
if isinstance(refine_template, str):
|
|
self._refine_template = PromptTemplate(refine_template)
|
|
else:
|
|
self._refine_template = refine_template or DEFAULT_REFINE_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 contexts is relevant to the query."""
|
|
del kwargs # Unused
|
|
del response # Unused
|
|
|
|
if query is None or contexts is None:
|
|
raise ValueError("Both query and contexts must be provided")
|
|
|
|
docs = [Document(text=context) for context in contexts]
|
|
index = SummaryIndex.from_documents(docs, service_context=self._service_context)
|
|
|
|
await asyncio.sleep(sleep_time_in_seconds)
|
|
|
|
query_engine = index.as_query_engine(
|
|
text_qa_template=self._eval_template,
|
|
refine_template=self._refine_template,
|
|
)
|
|
response_obj = await query_engine.aquery(query)
|
|
raw_response_txt = str(response_obj)
|
|
|
|
score, reasoning = self.parser_function(raw_response_txt)
|
|
|
|
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,
|
|
contexts=contexts,
|
|
score=score,
|
|
feedback=raw_response_txt,
|
|
invalid_result=invalid_result,
|
|
invalid_reason=invalid_reason,
|
|
)
|