Magpie-Align/Magpie-Air-300K-Filtered
收藏Hugging Face2024-07-03 更新2024-06-22 收录
下载链接:
https://hf-mirror.com/datasets/Magpie-Align/Magpie-Air-300K-Filtered
下载链接
链接失效反馈官方服务:
资源简介:
该数据集是通过Llama 3 8B Instruct模型使用Magpie方法生成的,包含300,000个高质量的对话实例。这些实例是从生成的4百万条指令中筛选出来的,筛选标准包括输入质量(>= good)、输入难度(>= medium)、奖励差异(>= 0)等。数据集用于监督微调(SFT)模型,并且在某些任务上,使用Magpie数据集微调的模型表现与官方Llama-3-8B-Instruct模型相当。
This dataset is generated by Llama 3 8B Instruct using the Magpie method, containing 300,000 high-quality conversation instances. These instances are filtered from 4 million generated instructions based on criteria such as input quality (>= good), input difficulty (>= medium), reward difference (>= 0), etc. The dataset is used for supervised fine-tuning (SFT) of models, and in some tasks, models fine-tuned with the Magpie dataset perform comparably to the official Llama-3-8B-Instruct model.
提供机构:
Magpie-Align
原始信息汇总
数据集概述
数据特征
- conversations: 包含对话信息
- from: 字符串类型
- value: 字符串类型
- uuid: 字符串类型
数据分割
- train: 训练集
- num_bytes: 972992281 字节
- num_examples: 300000 条样本
数据大小
- download_size: 499570324 字节
- dataset_size: 972992281 字节
配置
- default: 默认配置
- data_files:
- split: train
- path: data/train-*
- data_files:



