Qian et al.

Researchers Qian et al. introduced ToolRL, a reinforcement learning method for tool use that uses a decomposed reward function—format plus correctness—outperforming supervised fine-tuning imitation. A…