Back to Articles
TL;DR
I trained lattice-retrieval, a static embedding model, on 660M curated
query/document pairs. It scores 0.4581 NDCG@10 on decontaminated BEIR before
fine-tuning and 0.4749 after fine-tuning, compared with 0.4334 for
Lattice: an 8 MB static retriever that embeds Wikipedia in 7 minutes
A Blog post by Erik Kaunismäki on Hugging Face
4,297 words~20 min read
