The embedding models are already defined, but they aren't ever actually registered. This means they can't be used with llm embed. For example:

$ llm embed -m github/text-embedding-3-small -c "can my pup eat brassicas?"
Error: You need to specify an embedding model (no default model is set)

The models can be registered via the register_embedding_models llm hook and there's an llm.EmbeddingModel class which provides the basic support.