retriever/vectordb_rerank.py

# vectordb_relank_law.py
import faiss
import numpy as np
import os
from chromadb import PersistentClient
from chromadb.utils import embedding_functions
from sentence_transformers import SentenceTransformer
from retriever.reranker import rerank_documents

# chroma vector config v2
embedding_models = [
    "upskyy/bge-m3-korean", 
    "jhgan/ko-sbert-sts", 
    "BM-K/KoSimCSE-roberta", 
    "BM-K/KoSimCSE-v2-multitask", 
    "snunlp/KR-SBERT-V40K-klueNLI-augSTS", 
    "beomi/KcELECTRA-small-v2022", 
]
# law_db config v2
CHROMA_PATH = os.path.abspath("data/index/law_db")
COLLECTION_NAME = "law_all"
EMBEDDING_MODEL_NAME = embedding_models[0]  # 사용하고자 하는 모델 선택


# 1. 임베딩 모델 로드 v2
# embedding_model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
embedding_model = SentenceTransformer(EMBEDDING_MODEL_NAME)


# 2. 임베딩 함수 설정
embedding_fn = embedding_functions.SentenceTransformerEmbeddingFunction(model_name=EMBEDDING_MODEL_NAME)

# 3. Chroma 클라이언트 및 컬렉션 로드
client = PersistentClient(path=CHROMA_PATH)
collection = client.get_collection(name=COLLECTION_NAME, embedding_function=embedding_fn)



# 4. 검색 함수
def search_documents(query: str, top_k: int = 5):
    print(f"\n🔍 검색어: '{query}'")
    results = collection.query(
        query_texts=[query],
        n_results=top_k,
        include=["documents", "metadatas", "distances"]
    )

    for i, (doc, meta, dist) in enumerate(zip(
        results['documents'][0],
        results['metadatas'][0],
        results['distances'][0]
    )):
        print(f"\n📄 결과 {i+1} (유사도: {1 - dist:.2f})")
        print(f"문서: {doc[:150]}...")
        print("메타데이터:")
        print(meta)