performance optimization ideas

[jinshuaiwang]

when run the project with deepseek，document retrieve and rag invoke method costs more than 20 seconds，chat mode is graph + vector + fulltext，any performance optimization idea？