Those embedding models are used for semantic search, and don't work the same way, so as of now, you have to generate those embeddings yourself, sadly. A recommendation generation feature is in the works though!