llm / src /rag /vector_store.py
lotran12's picture
Upload 8 files
a6e314f verified
Raw History Blame Contribute Delete
1.68 kB
from typing import Union
from langchain_chroma import Chroma
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
class VectorDB:
def __init__(
self,
vector_db_cls: Union[Chroma, FAISS] = Chroma,
embedding=None,
) -> None:
self.vector_db_cls = vector_db_cls
self.embedding = embedding
self.db = None
def _build_db(self, documents):
return self.vector_db_cls.from_documents(
documents=documents,
embedding=self.embedding
)
def get_retriever(self,
search_type: str = "similarity",
search_kwargs: dict = {"k": 10}):
if 'k' not in search_kwargs:
search_kwargs['k'] = 10
if self.db is not None:
retriever = self.db.as_retriever(search_type=search_type,
search_kwargs=search_kwargs)
else:
retriever = None
return retriever
def update_db(self, new_documents):
if not new_documents:
raise ValueError("New documents must be provided to update the database.")
if self.db is None:
self.db = self._build_db(new_documents)
else:
self.db.add_documents(new_documents)
def clear_vectors(self):
if self.db is None:
raise ValueError("Database is empty, cannot clear vectors.")
self.db = None
self.db = self.vector_db_cls.from_documents(
documents=[],
embedding=self.embedding
)
print("All vectors have been cleared.")