arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.02160cs.CL
收藏

D-CORE:激励大型推理模型在复杂工具使用中的任务分解

D-CORE: Incentivizing Task Decomposition in Large Reasoning Models for Complex Tool Use

Alibaba Cloud Computing, Alibaba Group(阿里巴巴云 computing,阿里巴巴集团)

Bowen Xu, Shaoyu Wu, Hao Jiang, Kai Liu, Xin Chen, Lulu Hu, Bin Yang

AI总结:

D-CORE通过两阶段训练框架提升大型推理模型在复杂工具使用中的任务分解能力,实验显示其在多个基准上优于现有方法。

摘要:

有效的工具使用和推理能力是大型推理模型(LRMs)解决复杂现实问题所必需的核心能力。通过实证分析,我们发现当前LRMs在复杂工具使用场景中缺乏子任务分解能力,导致懒惰推理。为此,我们提出了一种两阶段训练框架D-CORE(分解任务和组合推理过程),首先通过自我蒸馏激励LRMs的任务分解推理能力,随后通过多样性感知强化学习(RL)恢复LRMs的反思推理能力。D-CORE在多样化的基准和模型规模上实现了稳健的工具使用改进。在BFCLv3上的实验表明,我们的方法具有优势:D-CORE-8B达到77.7%的准确率,比最佳表现的8B模型高出5.7%。同时,D-CORE-14B在70B模型之上建立了新的最先进的水平,尽管其规模小5倍。源代码可在https://github.com/alibaba/EfficientAI上获得。

英文摘要:

Effective tool use and reasoning are essential capabilities for large reasoning models~(LRMs) to address complex real-world problems. Through empirical analysis, we identify that current LRMs lack the capability of sub-task decomposition in complex tool use scenarios, leading to Lazy Reasoning. To address this, we propose a two-stage training framework D-CORE~(\underline{\textbf{D}}ecomposing tasks and \underline{\textbf{Co}}mposing \underline{\textbf{Re}}asoning processes) that first incentivize the LRMs' task decomposition reasoning capability via self-distillation, followed by diversity-aware reinforcement learning~(RL) to restore LRMs' reflective reasoning capability. D-CORE achieves robust tool-use improvements across diverse benchmarks and model scales. Experiments on BFCLv3 demonstrate superiority of our method: D-CORE-8B reaches 77.7\% accuracy, surpassing the best-performing 8B model by 5.7\%. Meanwhile, D-CORE-14B establishes a new state-of-the-art at 79.3\%, outperforming 70B models despite being 5$\times$ smaller. The source code is available at https://github.com/alibaba/EfficientAI.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑