让大模型学会拆解复杂任务，提升工具使用能力
D-CORE: Incentivizing Task Decomposition in Large Reasoning Models for Complex Tool Use
- 通过自蒸馏和强化学习，强制模型分解复杂任务
- 在BFCLv3上8B模型达77.7%准确率，超最佳8B模型5.7%
- 14B模型达79.3%超越70B模型，适合小模型高效部署
大型推理模型（LRMs）有效使用工具与推理能力对解决现实复杂问题至关重要。我们通过实证分析发现，当前LRMs在复杂工具使用场景中缺乏子任务分解能力，导致惰性推理。为此，我们提出两阶段训练框架D-CORE（分解任务与组合推理过程），首先通过自蒸馏激励模型的任务分解能力，再利用多样性感知强化学习恢复其反思性推理能力。D-CORE在多个基准和模型规模下均实现稳健的工具使用性能提升。在BFCLv3上的实验表明：D-CORE-8B达到77.7%准确率，超越最佳8B模型5.7%；D-CORE-14B达79.3%，虽仅是70B模型的五分之一，却已刷新当前最优表现。代码已开源。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。