Each entry specifies the HuggingFace transformer model, its embedding
dimension, and the corresponding MLP checkpoint filenames distributed
via piggyback. Users wanting to plug in a different sentence-transformer
(e.g., one of the newer top entries on
MTEB) can bypass
this registry by passing a list directly to ebisg.model with their
own transformer, dim, and surname_mlp (path to a local
.pt checkpoint). See resolve_ebisg_model() for the schema.
ebisg_model_registryAn object of class list of length 1.