TIGER-Lab/Fineweb-Instruct
收藏官方服务：
资源简介：
Fineweb-Instruct数据集是从Fineweb-Edu预训练语料库转换而来的，采用了指令跟随格式。通过质量筛选后，使用GPT-4提取了约1600万条指令对。数据集的基本概念类似于MAmmoTH2。
The Fineweb-Instruct dataset is converted from the Fineweb-Edu pre-training corpus into an instruction following format. A subset was selected with quality filtering and instruction-following pairs were extracted using GPT-4. The dataset contains roughly 16 million instruction pairs, with a basic concept similar to MAmmoTH2.
提供机构：
TIGER-Lab
