DICE-BENCH
收藏arXiv2025-09-30 收录
下载链接：
https://snuhcc.github.io/DICE-Bench/
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为DICE-BENCH，它通过构建一个工具图来合成对话，这个工具图能够在多轮对话中维持依赖关系，并结合一个具有不同角色的多代理系统，以增强对话的自然度。该数据集包含高DICE-SCORE实例，旨在评估大型语言模型在现实多轮对话中的表现。其规模为1,607个实例，任务是对大型语言模型在多轮、多方对话中使用工具的能力进行评估。
The dataset is named DICE-BENCH. It synthesizes dialogues by constructing a tool graph that maintains dependencies across multi-turn conversations, and incorporates a multi-agent system with diverse roles to enhance the naturalness of the dialogues. This dataset includes high DICE-SCORE instances, which is designed to evaluate the performance of large language models (LLMs) in realistic multi-turn conversations. It comprises 1,607 instances in total, with the core task of assessing the tool-use capabilities of large language models in multi-turn, multi-party conversations.
