KURE rerankers demoed: 149M nano model boosts the right answer to top spot
tomaarsen · x · 2026-10-23
tomaarsen, maintainer of Sentence Transformers, shows KURE reranking models: a 1.7B base and a 149M nano, both usable via CrossEncoder. A demo ranks candidate answers to a photosynthesis query, with the correct passage scoring highest (-3.27). Scores are relevance logits — higher is better even when negative.
More from coding & agent
- Open-source tool turns the MacBook notch into a one-click approval panel for Claude Code — cneuralnetwork · 2026-10-23
- Taskuary: open-source bridge that routes email and chat tasks to Codex, Claude Code and Gemini CLIs — Appropriate-Path-461 · 2026-10-23
- AI makes coding 10x faster, so why isn't your team shipping 10x faster? Amdahl's law explains — arpit_bhayani · 2026-10-23
- Side project builds textual 3D worlds in the terminal with custom TUI tooling — sull · 2026-10-23
- Selfdrive AI agent turns an App Store link into a full user-acquisition plan — PrajwalTomar_ · 2026-10-23
- Thorsten Ball mocks SaaS firms building internal agent platforms nobody uses — HankYeomans · 2026-10-23