[gdj0nes]

We trained a sparse embedding model because there is a lack of commercially licensed weights or APIs. For any proprietary model, particularly embeddings, there is a degree of lock-in.