A vector store retriever is a component or module that specializes in retrieving vectors from a vector store based on user queries. It works in conjunction with a vector store to facilitate efficient vector retrieval and similarity search operations.
Note on dissimilarity scores: Dissimilarity scores calculated using FAISS or Chroma with L2 distance are not bound to a specific range, therefore allowing only for ordinal comparison of scores. These scores also depend on the embeddings model used to generate the embeddings, as different models produce embeddings with varying scales and distributions. Therefore, understanding or comparing similarity across different models or spaces without contextual normalization is not meaningful.