init

apps/agent/multi_query_chain.py

@@ -33,7 +33,5 @@ QUERY_PROMPT = PromptTemplate(
     input_variables=["question"],
     template=template,
 )
-# llm = ChatOpenAI(temperature=0, openai_api_key=OPENAI_API_KEY)
-
 # Chain
 llm_chain = QUERY_PROMPT | llm | output_parser

apps/agent/tools.py

@@ -16,13 +16,14 @@ compressor = FlashrankRerank(model="ms-marco-MiniLM-L-12-v2")
 def multiquery_retriever(index_name: str, embeddings, compressor) -> ContextualCompressionRetriever:
     vectorstore = Pinecone.from_existing_index(embedding=embeddings, index_name=index_name)
     retriever = vectorstore.as_retriever()
-    reranker_retriever = ContextualCompressionRetriever(
-        base_compressor=compressor, base_retriever=retriever
+    
+    multi_retriever = MultiQueryRetriever(
+        retriever=retriever, llm_chain=llm_chain, parser_key="lines"
     )
-    print("Initialize MultiQuery...")
-    return MultiQueryRetriever(
-        retriever=reranker_retriever, llm_chain=llm_chain, parser_key="lines"
+    reranker_retriever = ContextualCompressionRetriever(
+        base_compressor=compressor, base_retriever=multi_retriever
     )
+    return reranker_retriever
 
 retriever_xano = multiquery_retriever(INDEX_NAME_XANO, embeddings, compressor)
 retriever_weweb = multiquery_retriever(INDEX_NAME_WEWEB, embeddings, compressor)