Keyword Search Matches 8B Embedding Model in Agent Tool Selection

On 200 ToolRet questions with up to 3,794 tools, BM25 keyword search ran about 40x faster than BGE-large on a CPU while matching an 8B embedding model, with full tool docs yielding the best tested setup.

Read the full story at Towards Data Science ↗

Timeline · 1 report

2026-11-13 16:30 · Towards Data Science
Keyword Search Tied an 8B Embedding Model at Picking Agent Tools

More stories

Jensen Huang resigns as Nvidia CEO amid spy scandal and chip smuggling probes — CSET Georgetown
OpenAI launches ChatGPT Wallet for agent purchases — OpenAI News
Google launches Gemini 4 Hydrogen, a faster model for coding and Search — Google AI Blog
Jensen Huang steps down as Nvidia CEO, sending shockwaves through AI markets — Euronews Next
Google Wallet connects to Gemini for personal finance insights in the U.S. — Google AI Blog
Ai2 launches XYBench to test whether LLMs spot misconceptions behind user questions — Allen Institute for AI (Ai2)
OpenAI board member Paul Christiano warns AI accident is most likely cause of death — Gizmodo AI
Democratic group Tomorrow Together launches AI agenda push ahead of 2028 election — Washington Post AI

Get the daily brief of stories like this at 6:30 every morning →