reorder program

app.py

@@ -29,13 +29,6 @@ ZIP_FILE = "chroma_db.zip"
 reranker_model = None
 embedding = None
 
-if not os.path.exists(CHROMA_PATH):
-    print("Downloading ChromaDB from Google Drive...")
-    subprocess.run(["gdown", f"https://drive.google.com/uc?id={GOOGLE_DRIVE_FILE_ID}", "-O", ZIP_FILE])
-    subprocess.run(["unzip", ZIP_FILE])  # Extract database
-    retriever = Chroma(persist_directory=CHROMA_PATH, embedding_function=embedding).as_retriever(search_kwargs={"k": 20})
-    print("ChromaDB loaded!")
-
 def load_embedding_model(model_path : str):
     start_time = time.time()
     encode_kwargs = {"normalize_embeddings": True}
@@ -48,6 +41,21 @@ def load_embedding_model(model_path : str):
     print(f'model load time {round(end_time - start_time, 0)} second')
     return local_embedding
 
+if not reranker_model:
+    reranker_model = HuggingFaceCrossEncoder(model_name="BAAI/bge-reranker-v2-m3")
+    print("reranker model loaded")
+if not embedding:
+    embedding = load_embedding_model(model_path="intfloat/multilingual-e5-large")
+    print("embedding model loaded")
+
+if not os.path.exists(CHROMA_PATH):
+    print("Downloading ChromaDB from Google Drive...")
+    subprocess.run(["gdown", f"https://drive.google.com/uc?id={GOOGLE_DRIVE_FILE_ID}", "-O", ZIP_FILE])
+    subprocess.run(["unzip", ZIP_FILE])  # Extract database
+    retriever = Chroma(persist_directory=CHROMA_PATH, embedding_function=embedding).as_retriever(search_kwargs={"k": 20})
+    print("ChromaDB loaded!")
+
+
 def rag_with_reranking(query : str):
     compressor = CrossEncoderReranker(model=reranker_model, top_n=3)
     compression_retriever = ContextualCompressionRetriever(
@@ -59,15 +67,8 @@ def rag_with_reranking(query : str):
 @app.get("/search")
 def search_text(query):
     """Searches for similar texts."""
-    global reranker_model, embedding
-    
+
     print(f"Searching for: {query}")
-    if not reranker_model:
-        reranker_model = HuggingFaceCrossEncoder(model_name="BAAI/bge-reranker-v2-m3")
-        print("reranker model loaded")
-    if not embedding:
-        embedding = load_embedding_model(model_path="intfloat/multilingual-e5-large")
-        print("embedding model loaded")
 
     results = rag_with_reranking(query)