integrated milvus db

application/core/settings.py

@@ -18,7 +18,7 @@ class Settings(BaseSettings):
     DEFAULT_MAX_HISTORY: int = 150
     MODEL_TOKEN_LIMITS: dict = {"gpt-3.5-turbo": 4096, "claude-2": 1e5}
     UPLOAD_FOLDER: str = "inputs"
-    VECTOR_STORE: str = "faiss"  # "faiss" or "elasticsearch" or "qdrant"
+    VECTOR_STORE: str = "faiss"  # "faiss" or "elasticsearch" or "qdrant" or "milvus"
     RETRIEVERS_ENABLED: list = ["classic_rag", "duckduck_search"] # also brave_search
 
     API_URL: str = "http://localhost:7091"  # backend url for celery worker
@@ -61,6 +61,11 @@ class Settings(BaseSettings):
     QDRANT_PATH: Optional[str] = None
     QDRANT_DISTANCE_FUNC: str = "Cosine"
 
+    # Milvus vectorstore config
+    MILVUS_COLLECTION_NAME: Optional[str] = "docsgpt"
+    MILVUS_URI: Optional[str] = "./milvus_local.db"   # milvus lite version as default
+    MILVUS_TOKEN: Optional[str] = ""
+
     BRAVE_SEARCH_API_KEY: Optional[str] = None
 
     FLASK_DEBUG_MODE: bool = False

application/vectorstore/milvus.py

@@ -0,0 +1,38 @@
+from typing import List, Optional
+from langchain_community.vectorstores.milvus import Milvus
+
+from application.core.settings import settings
+from application.vectorstore.base import BaseVectorStore
+
+
+class MilvusStore(BaseVectorStore):
+    def __init__(self, path: str = "", embeddings_key: str = "embeddings"):
+        super().__init__()
+        if path:
+            connection_args ={
+                "uri": path,
+                "tpken": settings.MILVUS_TOKEN,
+            }
+        else:
+            connection_args = {
+                "uri": settings.MILVUS_URL,
+                'token': settings.MILVUS_TOKEN,
+            }
+        self._docsearch = Milvus(
+            embedding_function=self._get_embeddings(settings.EMBEDDINGS_NAME, embeddings_key),
+            collection_name=settings.COLLECTION_NAME,
+            connection_args=connection_args,
+            drop_old=True,
+        )
+
+    def search(self, question, k=2, *args, **kwargs):
+        return self._docsearch.similarity_search(query=question, k=k, *args, **kwargs)
+
+    def add_texts(self, texts: List[str], metadatas: Optional[List[dict]], *args, **kwargs):
+        return self._docsearch.add_texts(texts=texts, metadatas=metadatas, *args, **kwargs)
+
+    def save_local(self, *args, **kwargs):
+        pass
+
+    def delete_index(self, *args, **kwargs):
+        pass

application/vectorstore/vector_creator.py

@@ -1,5 +1,6 @@
 from application.vectorstore.faiss import FaissStore
 from application.vectorstore.elasticsearch import ElasticsearchStore
+from application.vectorstore.milvus import MilvusStore
 from application.vectorstore.mongodb import MongoDBVectorStore
 from application.vectorstore.qdrant import QdrantStore
 
@@ -10,6 +11,7 @@ class VectorCreator:
         "elasticsearch": ElasticsearchStore,
         "mongodb": MongoDBVectorStore,
         "qdrant": QdrantStore,
+        "milvus": MilvusStore,
     }
 
     @classmethod