We have a flag to signal we are shortening the embeddings of a model. Only used in Open AI's text-embedding-3-*, but we plan to use it for other services.
