Evol-Instruct Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/nlpxucan/WizardLM
下载链接
链接失效反馈官方服务：
资源简介：
该数据集最初由52,000个指令样本组成，经过迭代过程，利用大型语言模型（LLMs）将其扩展至250,000个指令。这一过程涉及多轮指令进化，并在LLaMA模型上进行微调。目前，该数据集的规模为250,000个指令，其任务是遵循指令。
Initially consisting of 52,000 instruction samples, this dataset was expanded to 250,000 instructions via an iterative process leveraging large language models (LLMs). This process entails multi-round instruction evolution and fine-tuning on the LLaMA model. Currently, the dataset comprises 250,000 instruction samples, and its core task is instruction following.
提供机构：
NLPXUCan
