Same technique as for mixed password hashing, you store the embedding model name and encode and search per each model stored, until the multiple embedding cost

Same technique as for mixed password hashing, you store the embedding model name and encode and search per each model stored, until the multiple embedding cost at search time become larger than embedding the old data again in the new for. At