MCQA Benchmark
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/matthewrenze/jhu-concise-cot
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个用于评估简洁思维链（CCoT）提示对大型语言模型回答长度和准确性的影响的多选题问答基准。此外，该数据集还用于比较在不同提示策略下GPT-3.5和GPT-4的性能，研究结果表明，这种策略能够减少回答长度并节省成本。该数据集包含了1,000个问题，任务类型为问答。
This dataset is a multiple-choice question answering benchmark developed to assess the impact of Concise Chain-of-Thought (CCoT) prompting on the response length and accuracy of Large Language Models (LLMs). Furthermore, this dataset is employed to compare the performance of GPT-3.5 and GPT-4 under various prompting strategies. Research findings demonstrate that this prompting strategy can reduce response length and lower operational costs. This dataset comprises 1,000 questions, with the task type being question answering.
