QReason paper decouples chain-of-thought to cut reranking latency without losing accuracy
_reachsumit · x · 2026-09-28
The arXiv paper QReason introduces a query-focused decoupled chain-of-thought framework for passage reranking. Listwise LLM rerankers with sliding windows regenerate near-identical CoTs repeatedly, adding redundancy and latency. QReason trains a rewriter to produce a ranking-oriented reasoning query once and reuse it across all windows with a non-reasoning reranker. Training is two-stage: SFT with relevant-passage semantic evidence, then RL aligning CoT generation with listwise metrics and passage discrimination. On the BRIGHT benchmark it significantly reduces redundant reasoning while matching or beating strong reasoning rerankers.
More from Research
- Dev rebuilds old paper into Synthetic Social Graph, an interactive social-dynamics energy-field simulator — seanmcdonaldxyz · 2026-10-24
- UC Berkeley's 'Modern Computer Vision' Course Lectures Now Freely Available Online — burny_tech · 2026-10-24
- New Deep Learning Framework Regresses Random Objects in Metric Spaces Without Manifold Assumptions — burny_tech · 2026-10-24
- Cell paper: in vivo gene writing with engineered retrotransposons — anshulkundaje · 2026-10-24
- Reddit deep-dive: next-gen AI via memristors, spiking nets, and 1% sparse routing — Robert__Sinclair · 2026-10-24
- FADA adapts humanoid robots to new conditions in ~2 minutes, no rewards needed — GuanyaShi · 2026-10-24