I have updated Vajra BM25 to a new minor version and it is now available on PyPI pypi.org/project/vajra-… and Github github.com/aiexplorations… The benchmark looks really good for 1 million documents, Wikipedia data set. The closest library that performs this way is BM25s. Vajra is trading off build (slower) for better query performance (latency). These are non-cached results. With an LRU cache, Vajra Search is massively faster, and results from warm caches are near instantaneous. Vajra BM25 is arguably the fastest full text BM25 search now built on top of Python, comparable to BM25S. Learn more here on the blog post: rajeshrs.in/blog/ai-explor…
