ColGrep-Agent: a 2B-parameter local semantic-search agent matches 10x larger models on code localization
IgorCarron · x · 2026-10-12
LightOn released a paper (arXiv 2610.05099) and open-source tooling showing small specialized agents can handle file localization for coding agents. Their recipe combines ColGREP (local late-interaction semantic search), weighted SFT on teacher trajectories from a 27B model, and GRPO-based RL on localization quality. Models under 2B parameters hit 76.18 file F1 on SWE-bench Lite and 62.57 on Multi-SWE-bench Flash, matching models 10x their size, while cutting CPU end-to-end latency by 44.1% and tokens by 29.1%, with better cross-language generalization.
More from coding & agent
- Rigging in Blender via MCP: six-step recipe for Opus 5.5 to build and animate 3D scenes — sidahuj · 2026-10-12
- Diorama gives each agent its own git worktree — and bangs a giant gong when stuck — davidfromkansas · 2026-10-12
- Tracking people across home security cameras with cheap frame sampling and Haiku 5.5 — indragie · 2026-10-12
- How the cat dashboard works: Ring's AI tagging plus 1fps still frames fed to the model — indragie · 2026-10-12
- He built a cat activity dashboard with Opus 5.5 by sniffing Ring, Petlibro and Whisker app traffic — indragie · 2026-10-12
- Diorama breaks down agent work into abstraction layers so you can see what's happening — davidfromkansas · 2026-10-12