VectorStoreIndex as retriever has faster LLM generation as compared to using a vector database like chroma. This is noticeable specially if you are running open-source LLM like Llama 2. Based on this observation I am likely to drop chroma out of the pipeline. Can someone change my mind?Chroma is a Vector Database, used to store the vector/embeddings at a place other than local. So that you can access the embeddings from any place or any system.Chroma, You can put them into VectorStoreIndex and it will work the same as normal VectorStoreIndex.