TreeHCA

TreeHCA, a new reinforcement-learning method for training multi-turn LLM agents, outperformed all baselines in aggregate exact match across seven question-answering benchmarks and three Qwen backbones…