The authors address the bottleneck of tool selection in LLM agents when tool corpora exceed context limits.
Small language models trained with reinforcement learning outperform 402B model in tool selection
A Qwen3-8B model beats Llama-4-Maverick on ToolBench and transfers zero-shot to new tool corpora and retrievers.
Top University
Zhaoyi Zhou · Akshayaa Magesh · Jeongyeol Kwon · Artem Kuriksha · Daniel Jiang · Andrea Zanette · +1 more
Meta · Carnegie Mellon University
Research Digest··2 min read
7B, 4B, 8B) using reinforcement learning to select tools for LLM agents over large tool corpora.
§