Salesforce/MASBench

MASBench数据集是MAS-Orchestra项目的一部分，旨在通过整体编排和受控基准来理解和改进多智能体推理。该数据集用于文本生成任务，包含多个配置（广度、组合、深度、水平、并行、鲁棒性），每个配置具有不同的特征，如reward_model_json、extra_info_json、prompt_json、axis、value和data。每个配置都有训练和测试分割，并指定了大小和示例数量。数据集采用Apache-2.0许可证，仅用于研究目的。

The MASBench dataset is part of the MAS-Orchestra project, which aims to understand and improve multi-agent reasoning through holistic orchestration and controlled benchmarks. The dataset is used for text-generation tasks and includes multiple configurations (breadth, combine, depth, horizon, parallel, robustness) with various features such as reward_model_json, extra_info_json, prompt_json, axis, value, and data. Each configuration has train and test splits with specified sizes and examples. The dataset is licensed under Apache-2.0 and is intended for research purposes only.