Maybe I need to change embedding model?

            query_engine = self.index.as_query_engine(service_context=self.ctx, streaming=True)

self.index = VectorStoreIndex.from_documents(documents, service_context=self.ctx)
# or
self.index = load_index_from_storage(storage_context, service_context=self.ctx)

def _initialize_index(self):
    self.ctx = ServiceContext.from_defaults(llm=llm, embed_model="local")
    if not os.path.exists(self.persist_dir):
        documents = SimpleDirectoryReader("data").load_data()
        LLAMA_13B_V2_CHAT = "a16z-infra/llama13b-v2-chat:df7690f1994d94e96ad9d568eac121aecf50684a0b0963b25a41cc40061269e5"
        llm = Replicate(
            model=LLAMA_13B_V2_CHAT,
            temperature=0.01,
            context_window=4096,
            completion_to_prompt=self.custom_completion_to_prompt,
            messages_to_prompt=messages_to_prompt,
        )
        self.index = VectorStoreIndex.from_documents(documents, service_context=self.ctx)
        self.index.storage_context.persist(persist_dir=self.persist_dir)
    else:
        storage_context = StorageContext.from_defaults(persist_dir=self.persist_dir)
        self.index = load_index_from_storage(storage_context, service_context=self.ctx)