This model (extreme learning machine, a shallow neural net) was trained in R, to reproject sentence transformers ('all-mpnet-base-v2') into wikidata5m knowledge graph embeddings (rotate version).
It is stored with fastsave (
https://github.com/barkasn/fastSave_, depends on the library elmNNRcpp
(
https://cran.r-project.org/web/packages/elmNNRcpp/index.html).
It was trained on a subset of the knowledge graph embeddings (400k version) of entities, which matched with the common sense knowledge graph or the entities in numberbatch.
I intend to redo this with a larger subset / full data.
