mirror of
https://github.com/hwchase17/langchain.git
synced 2025-05-16 12:32:06 +00:00
c4d2a53f18
Description: add a optional score relevance threshold for select only coherent document, it's in complement of top_n Discussion: add relevance score threshold in flashrank_rerank document compressors #24013 Dependencies: no dependencies --------- Co-authored-by: Benjamin BERNARD <benjamin.bernard@openpathview.fr> Co-authored-by: Bagatur <22008038+baskaryan@users.noreply.github.com> Co-authored-by: Chester Curme <chester.curme@gmail.com>
87 lines
2.8 KiB
Python
87 lines
2.8 KiB
Python
from __future__ import annotations
|
|
|
|
from typing import TYPE_CHECKING, Dict, Optional, Sequence
|
|
|
|
from langchain_core.callbacks.manager import Callbacks
|
|
from langchain_core.documents import BaseDocumentCompressor, Document
|
|
from langchain_core.pydantic_v1 import Extra, root_validator
|
|
|
|
if TYPE_CHECKING:
|
|
from flashrank import Ranker, RerankRequest
|
|
else:
|
|
# Avoid pydantic annotation issues when actually instantiating
|
|
# while keeping this import optional
|
|
try:
|
|
from flashrank import Ranker, RerankRequest
|
|
except ImportError:
|
|
pass
|
|
|
|
DEFAULT_MODEL_NAME = "ms-marco-MultiBERT-L-12"
|
|
|
|
|
|
class FlashrankRerank(BaseDocumentCompressor):
|
|
"""Document compressor using Flashrank interface."""
|
|
|
|
client: Ranker
|
|
"""Flashrank client to use for compressing documents"""
|
|
top_n: int = 3
|
|
"""Number of documents to return."""
|
|
score_threshold: float = 0.0
|
|
"""Minimum relevance threshold to return."""
|
|
model: Optional[str] = None
|
|
"""Model to use for reranking."""
|
|
prefix_metadata: str = ""
|
|
"""Prefix for flashrank_rerank metadata keys"""
|
|
|
|
class Config:
|
|
"""Configuration for this pydantic object."""
|
|
|
|
extra = Extra.forbid
|
|
arbitrary_types_allowed = True
|
|
|
|
@root_validator(pre=True)
|
|
def validate_environment(cls, values: Dict) -> Dict:
|
|
"""Validate that api key and python package exists in environment."""
|
|
if "client" in values:
|
|
return values
|
|
else:
|
|
try:
|
|
from flashrank import Ranker
|
|
except ImportError:
|
|
raise ImportError(
|
|
"Could not import flashrank python package. "
|
|
"Please install it with `pip install flashrank`."
|
|
)
|
|
|
|
values["model"] = values.get("model", DEFAULT_MODEL_NAME)
|
|
values["client"] = Ranker(model_name=values["model"])
|
|
return values
|
|
|
|
def compress_documents(
|
|
self,
|
|
documents: Sequence[Document],
|
|
query: str,
|
|
callbacks: Optional[Callbacks] = None,
|
|
) -> Sequence[Document]:
|
|
passages = [
|
|
{"id": i, "text": doc.page_content, "meta": doc.metadata}
|
|
for i, doc in enumerate(documents)
|
|
]
|
|
|
|
rerank_request = RerankRequest(query=query, passages=passages)
|
|
rerank_response = self.client.rerank(rerank_request)[: self.top_n]
|
|
final_results = []
|
|
|
|
for r in rerank_response:
|
|
if r["score"] >= self.score_threshold:
|
|
doc = Document(
|
|
page_content=r["text"],
|
|
metadata={
|
|
self.prefix_metadata + "id": r["id"],
|
|
self.prefix_metadata + "relevance_score": r["score"],
|
|
**r["meta"],
|
|
},
|
|
)
|
|
final_results.append(doc)
|
|
return final_results
|