from typing import Any, List, Optional
from src.core.settings import Settings
from src.libs.llm.llm_factory import LLMFactory
from src.libs.reranker.base_reranker import BaseReranker
from src.libs.reranker.cross_encoder_reranker import CrossEncoderReranker
from src.libs.reranker.llm_reranker import LLMReranker
class NoneReranker(BaseReranker):
"""
No-op reranker that returns candidates in their original order.
Used as a fallback or when reranking is disabled.
"""
def rerank(
self,
query: str,
candidates: List[Any],
top_k: Optional[int] = None,
trace: Optional[Any] = None,
) -> List[Any]:
"""Return candidates as-is, optionally truncated to top_k."""
if top_k is not None:
return candidates[:top_k]
return candidates
class RerankerFactory:
"""Factory for creating Reranker instances based on configuration."""
@staticmethod
def create(settings: Settings) -> BaseReranker:
"""
Create a Reranker instance.
Args:
settings: Global settings object containing rerank configuration.
Returns:
An instance of BaseReranker.
Raises:
ValueError: If the configured backend is not supported.
"""
backend = settings.rerank.backend.lower()
if backend == "none":
return NoneReranker()
if backend == "llm":
llm = LLMFactory.create(settings)
return LLMReranker(llm)
if backend == "cross-encoder" or backend == "cross_encoder":
model_name = settings.rerank.model
return CrossEncoderReranker(model_name=model_name)
raise ValueError(f"Unsupported reranker backend: {backend}")
