Tag
The paper proposes DATPO, a difficulty-adaptive tree-structured policy optimization method that enhances reasoning coverage in large models through sentence-entropy-guided branching and diversity-aware optimization, outperforming baselines in pass@k metrics.