sangeet01/limitnumen

Training-free dense retrieval via character n-gram hashing. First dense model to beat BM25 on DeepMind's LIMIT benchmark. No training data required.

1 best points · 0 comments · 1 Hacker News post · 1 GitHub stars · View on GitHub

Hacker News discussions

• Show HN: Beating SOTA embeddings on DeepMind's LIMIT benchmark (94% vs. 18%) (1 points, 0 comments, by sangeet01, 2025-12-22)