CCCCCC/SPaR
收藏Hugging Face2024-10-30 更新2024-12-14 收录
下载链接：
https://hf-mirror.com/datasets/CCCCCC/SPaR
下载链接
链接失效反馈官方服务：
资源简介：
SPaR数据集旨在通过自我对弈框架提高语言模型的指令跟随能力。该数据集包含8,000个样本的SFT数据集，以及基于llama-3-8b-instruct和mistral-7b-instruct的DPO数据集。数据集主要用于指令跟随任务，数据主要为英文，原始提示来源于Infinity-Instruct数据集。
The SPaR Dataset is a self-play framework designed to enhance the instruction-following abilities of language models by generating high-quality preference pairs to minimize interfering factors. The dataset contains 8,000 samples curated using gpt-4o-mini, and provides DPO datasets derived from llama-3-8b-instruct and mistral-7b-instruct. It is primarily used for instruction-following tasks, particularly suitable for enhancing foundational instruction-following capabilities and preference learning. The data in the dataset is mostly in English.
提供机构：
CCCCCC
