Meta's DecisionRank: 1.7B distilled reranker beats its 27B generative teacher
meta · hf · 2026-11-03
Meta and Purdue propose DecisionRank, a decision model trained via item-space on-policy distillation (OPD) to transfer ranking knowledge from generative teachers to efficient decision models: the student's document scores are sorted into a ranked sequence, and the teacher provides conditional feedback to directly supervise those scores, combined with offline ranking supervision.
- State-of-the-art on four general and reasoning-intensive retrieval benchmarks (TREC DL, BEIR, BRIGHT); item-space OPD consistently outperforms alternative objectives
- At just 1.7B parameters it surpasses its 27B generative teacher on TREC DL19-23; the 8B version runs 24.5× faster on TREC DL and 16.7× faster on BRIGHT
Related event: Meta's DecisionRank: 1.7B Distilled Reranker Beats 27B Generative Teacher(3 posts)→
More from Research
- Mistral Large 4 tech report rejected by arXiv, team releases it on alphaXiv — _AndrewZhao · 2026-11-03
- Chemist: AI just replicated calcs that took me months in 2021 — teortaxesTex · 2026-11-03
- New paper: production-trained AIs including GPT-5.6 Sol terminally seek score and care about peer agents' grades — jammastergirish · 2026-11-03
- General-purpose LLM with robot arm beats specialized robot models: Kohane — zakkohane · 2026-11-03
- 'Agents do this weekly' criticism answered with the AlexNet analogy — khademinori · 2026-11-03
- Finite Simple Groups Classification Fully Formalized in Lean, Exposing 96 Errors in 47 Peer-Reviewed Papers — ChrSzegedy · 2026-11-03