The first dense retriever that can be prompted like a language model: trained on a released MS MARCO variant augmented with ~500K instruction-annotated instances, it follows free-form relevance instructions at query time and gains robustness to query phrasing while staying competitive on standard retrieval. JHU (Weller, Van Durme et al.) with Samaya AI and UW co-authors — note the joint attribution.

Paper

nlpopen-weightresearch

Related