DeepPlanning
收藏资源简介:
DeepPlanning 是一个用于评估大型语言模型(LLMs)在具有可验证约束条件下的长视野代理规划能力的挑战性基准。该数据集包含现实世界的多日旅行规划和多产品购物任务,要求模型具备主动信息获取、局部约束推理和全局约束优化能力。数据集主要涵盖两个领域:旅行规划(涉及时间、地点和预算的紧密耦合约束)和购物规划(需要在最大化折扣效用的同时找到最优产品的组合优化问题)。DeepPlanning 旨在填补当前代理评估中缺乏全局约束优化能力的空白,适用于文本生成、规划和推理等任务。数据集规模介于1k到10k之间,支持英文和中文,采用Apache-2.0许可。
DeepPlanning is a challenging benchmark for evaluating the long-horizon agent planning capabilities of large language models (LLMs) under verifiable constraints. This dataset includes real-world multi-day travel planning and multi-product shopping tasks, which demand the model to have capabilities including proactive information acquisition, local constraint reasoning, and global constraint optimization. The dataset covers two primary domains: travel planning, which involves tightly coupled constraints of time, location, and budget, and shopping planning, a combinatorial optimization problem that requires identifying the optimal product combination while maximizing discount utility. DeepPlanning aims to fill the gap in current agent evaluation where global constraint optimization capabilities are lacking, and is suitable for tasks such as text generation, planning, and reasoning. The dataset has a scale ranging from 1k to 10k, supports both English and Chinese, and is licensed under the Apache-2.0 license.
DeepPlanning 数据集概述
数据集基本信息
- 数据集名称: DeepPlanning
- 语言: 英语 (en), 中文 (zh)
- 许可证: Apache License 2.0
- 任务类别: 文本生成
- 标签: 规划, 大语言模型基准测试, 推理, 自主智能体
- 数据规模: 1k < n < 10k
数据集简介
DeepPlanningBench 是一个用于评估大语言模型在具有可验证约束条件下的长程智能体规划能力的挑战性基准测试。其特点是包含现实的多日旅行规划和多产品购物任务,这些任务要求具备主动信息获取、局部约束推理和全局约束优化的能力。
核心特点
- 主动信息获取: 通过 API 调用主动收集信息以发现隐藏的环境状态。
- 局部约束推理: 满足步骤级别的逻辑和特定要求。
- 全局约束优化: 管理整体边界,如总预算上限和多日时间可行性。
主要任务领域
- 旅行规划: 包含紧密耦合的时间、地点和预算约束的多日行程组织。
- 购物规划: 在最大化折扣效用的同时寻找最优产品的组合优化问题。
相关资源
- 项目网站: https://qwenlm.github.io/Qwen-Agent/en/benchmarks/deepplanning/
- 论文: https://arxiv.org/abs/2601.18137
引用信息
如果使用本数据集,请引用以下论文: bibtex @article{deepplanning, title={DeepPlanning: Benchmarking Long-Horizon Agentic Planning with Verifiable Constraints}, author={ Yinger Zhang and Shutong Jiang and Renhao Li and Jianhong Tu and Yang Su and Lianghao Deng and Xudong Guo and Chenxu Lv and Junyang Lin }, journal={arXiv preprint arXiv:2601.18137}, year={2026} }




