pushpdeep/UltraFeedback-paired
收藏Hugging Face2023-10-19 更新2024-03-04 收录
下载链接:
https://hf-mirror.com/datasets/pushpdeep/UltraFeedback-paired
下载链接
链接失效反馈官方服务:
资源简介:
---
license: mit
task_categories:
- text-generation
language:
- en
size_categories:
- 100K<n<1M
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: question
dtype: string
- name: response_j
dtype: string
- name: response_k
dtype: string
splits:
- name: train
num_bytes: 946257493
num_examples: 318777
download_size: 228559429
dataset_size: 946257493
---
# UltraFeedback Paired
This is a processed version of the [`openbmb/UltraFeedback`](https://huggingface.co/datasets/openbmb/UltraFeedback). The following steps were applied:
- Create pairs `(response_j, response_k)` where j was rated better than k based on `overall_score`
- Sample all 6 pairs for each instruction in the original data
This dataset is useful for LLM alignment techniques(like DPO). The processing steps are in [this repository](https://huggingface.co/datasets/pushpdeep/UltraFeedback-paired/blob/main/Ultrafeedback_paired_version.ipynb
). The code is based on [this repository](https://huggingface.co/datasets/lvwerra/stack-exchange-paired).
提供机构:
pushpdeep
原始信息汇总
UltraFeedback Paired 数据集概述
基本信息
- 许可证:MIT
- 任务类别:文本生成
- 语言:英语
- 数据规模:100K<n<1M
配置
- 配置名称:default
- 数据文件:
- 分割:train
- 路径:data/train-*
数据集信息
- 特征:
- 名称:question
- 数据类型:string
- 名称:response_j
- 数据类型:string
- 名称:response_k
- 数据类型:string
- 名称:question
- 分割:
- 名称:train
- 字节数:946257493
- 样本数:318777
- 名称:train
- 下载大小:228559429
- 数据集大小:946257493
处理步骤
- 创建配对
(response_j, response_k),其中 j 根据overall_score评级优于 k - 对原始数据中的每个指令采样所有 6 对
应用场景
- 适用于 LLM 对齐技术(如 DPO)



