https://docs.llamaindex.ai/en/stable/

Can you share your code, how you doing it

query_engine = CitationQueryEngine.from_args(
index,
node_postprocessors=[rerank],
similarity_top_k=20,
llm=llm,
)

This should not trim the number of nodes. Is there anything else in node postprocessor other than rerank?

no nothing other than rerank

What vector store are you using? Is the top_n for your reranker still 10?

milvus with zilliz

oh got it

rerank is 10