Our paper about efficient ColBERT was published at NLPCC. We apply JPQ to ColBERT: a compact PQ index to improve efficiency and joint optimization to improve effectiveness. JPQ achieves 10x compression ratio for ColBERT's index. link.springer.com/chapter/10.100…

"Disentangled Modeling" was rejected by #WSDM2023 But it is the best work I have worked on. Not only substantial OOD improvement for dense retrieval, but also for uniCOIL, SPLADE, ColBERT, and reranker! We will soon update the arxiv paper to include all these exciting results!

Introducing Demonstrate–Search–Predict (𝗗𝗦𝗣), a framework for composing search and LMs w/ up to 120% gains over GPT-3.5. No more prompt engineering.❌ Describe a high-level strategy as imperative code and let 𝗗𝗦𝗣 deal with prompts and queries.🧵 arxiv.org/abs/2212.14024