Answers on Unnatural Instructions
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/orhonovich/unnatural-instructions
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了68,000个指令-输入-输出三元组，用于评估GPT-4与经过指令调优的模型之间的性能差距。此外，该数据集也被用来量化模型在遵循指令能力方面的性能差异，其任务是对模型进行评估。
This dataset comprises 68,000 instruction-input-output triplets. It serves two core purposes: first, to evaluate the performance gap between GPT-4 and instruction-tuned models; second, to quantify the performance differences of models in terms of their instruction-following capabilities, acting as an evaluation benchmark for model assessment.
提供机构：
text-davinci-002
