Meta trains 8B Qwen3 model to beat larger LLMs at tool selection
Small models trained specifically for tool selection can outperform much larger prompted LLMs on the routing step, so agent builders can cut cost and latency by replacing the proprietary LLM in their tool-selection pipeline. The selector's ability to work with different retrievers and unseen domains makes it practical to slot into existing RAG-style stacks. Teams building large-scale agent tooling should compare this approach before defaulting to a frontier model for routing.
Meta released work on tool selection for agents with more than 10,000 tools, where tool descriptions exceed context limits and retrievers rank candidates without knowing what an instruction needs. The team trained Qwen3 models on a Plan, Retrieve, Select workflow: the model splits instructions into substeps, a retriever returns candidates, and the model commits one tool per substep. Training used SFT on about 20K curated ToolBench instructions followed by RL with a reward scoring gold-tool coverage and rank. On ToolBench G3, the 8B model improved from 36.2% after SFT to 41.5% after RL, beating a prompted 402B-class Llama-4-Maverick. The trained selector transfers across different retrievers and new-domain tool corpora.
Great application of small language models by Meta.
They find that an 8B model picks tools better than a 402B model when trained for the job.
This work focuses on tool selection for agents with more than ten thousand tools.
At that scale, tool descriptions won't fit in context, and a retriever can only rank candidates without knowing which tools the instruction actually needs. Many systems hand this step to a proprietary LLM.
They train Qwen3 models on a Plan, Retrieve, Select workflow. The model splits the instruction into substeps, the retriever returns candidates for each one, and the model commits one tool per substep. Training is SFT on about 20K curated ToolBench instructions, then RL with a reward that scores how many gold tools are covered and at what rank.
On ToolBench G3, the 8B model goes from 36.2% after SFT to 41.5% after RL, above prompted Llama-4-Maverick.
The trained selector also works with different retrievers and on tool corpora from new domains.
https://t.co/PoKsL4iqkG
Chat with Paper: https://t.co/6GMrNnk4tL
- Artificial Analysis10-21 17:35原文
