[How to: generate multiple embeddings per document](https://python.langchain.com/docs/how_to/multi_vector): LLM should read this page when needing to retrieve documents using multiple vector embeddings per document, when working with long documents that need to be split into chunks, when using document summaries for retrieval. This page covers how to index documents using 1) document chunks, 2) summaries generated with an LLM, and 3) hypothetical questions generated with an LLM. It demonstrates the usage of the MultiVectorRetriever to retrieve parent documents based on vector embeddings of chunks/summaries/questions.

