1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
|
import os
from typing import List
from loguru import logger as log
from sentence_transformers import CrossEncoder
from rag.message import Message
from rag.retriever.rerank.abstract import AbstractReranker
from rag.retriever.vector import Document
class Reranker(metaclass=AbstractReranker):
def __init__(self) -> None:
self.model = CrossEncoder(os.environ["RERANK_MODEL"], device="cpu")
self.top_k = int(os.environ["RERANK_TOP_K"])
self.relevance_threshold = float(os.environ["RETRIEVER_RELEVANCE_THRESHOLD"])
def rerank_documents(self, query: str, documents: List[Document]) -> List[str]:
results = self.model.rank(
query=query,
documents=[d.text for d in documents],
return_documents=False,
top_k=self.top_k,
)
ranking = list(
filter(lambda x: x.get("score", 0.0) > self.relevance_threshold, results)
)
log.debug(
f"Reranking gave {len(ranking)} relevant documents of {len(documents)}"
)
return [documents[r.get("corpus_id", 0)] for r in ranking]
def rerank_messages(self, query: str, messages: List[Message]) -> List[Message]:
results = self.model.rank(
query=query,
documents=[m.content for m in messages],
return_documents=False,
top_k=self.top_k,
)
ranking = list(
filter(lambda x: x.get("score", 0.0) > self.relevance_threshold, results)
)
log.debug(
f"Reranking gave {len(ranking)} relevant chat messages of {len(messages)}"
)
return [messages[r.get("corpus_id", 0)] for r in ranking]
|
