#tool-calling

Looped Language Models for Compositional Tool Calling

Standard LLMs struggle with complex, multi-step tool calls. By implementing a 'looped' architecture that allows models to iteratively refine and execute tool sequences, performance on compositional tasks significantly improves.