WikiBELA

This hub hosts the BELA model from Facebook, slighlty modified to store entity information from Wikidata in a SQLITE database.

Files:

• model_wiki.ckpt: multilingual bi-encoder model trained on Wikipedia,
• index.txt: text file containing the Wikidata identifiers of 16,470,856 Wikidata entities,
• faiss.index: a vector database containing the embeddings of the entities in the index obtained from the bi-encoder model,
• knowledge_base.sqlite: a look-up table containing the Wikidata label and description in 7 european languages (en, it, fr, de, nl, sv, fi), the earliest available date, and its entity type of each entity in the index.

---

license: apache-2.0