如何调用我 chromaDB 向量库中训练的内容

I hope to write the table structure and business explanations of some fields into a document in the chromaDB vector library, so that it can retrieve these related contents as context to write SQL when asking questions, however, the similarity search calculated by the built-in agent_memory is relatively low, even with the Chinese environment embedding function, the similarity is only 0.6, in this case, the training documents written are basically ineffective, I want to know how to solve this problem? I hope that he can generate SQL using the DDL table structure or related explanations, and then judge the table structure in the database