A prebuilt, file-backed Wikidata index optimized for entity linking (resolve), facet / type checks (is_a), compositional queries, and multilingual semantic similarity — no external database required. Everything is mmap-friendly on-disk artifacts (FST, roaring bitmaps, CSR graphs, usearch vector shards).
Reader / builder code:
https://github.com/mrink68/depesche-wd-index (Rust crates + wd CLI + PyO3 bindings)
Source: official… See the full description on the dataset page:
https://huggingface.co/datasets/Somnia/depesche-wd-index.
