eduagarcia/dpo-mix-21k

---
language:
- en
license: mit
size_categories:
- 10K<n<100K
tags:
- distilabel
- synthetic
- dpo
dataset_info:
features:
- name: dataset
dtype: string
- name: chosen
list:
- name: content
dtype: string
- name: role
dtype: string
- name: rejected
list:
- name: content
dtype: string
- name: role
dtype: string
- name: chosen_rating
dtype: float32
- name: rejected_rating
dtype: float32
splits:
- name: train
num_bytes: 120528535.0
num_examples: 20250
- name: test
num_bytes: 4555903.0
num_examples: 750
download_size: 198089631
dataset_size: 125084438.0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
---

# DPO Mix 21K Dataset

Similar to the [argilla/dpo-mix-7k](https://hf.co/datasets/argilla/dpo-mix-7k), but with 21k samples:

Other diferences:

- [`argilla/distilabel-capybara-dpo-7k-binarized`](https://huggingface.co/datasets/argilla/distilabel-capybara-dpo-7k-binarized) changed to [`argilla/Capybara-Preferences`](https://huggingface.co/datasets/argilla/Capybara-Preferences)
- [`argilla/distilabel-intel-orca-dpo-pairs`](argilla/distilabel-intel-orca-dpo-pairs) scored filter of >=8 changed to >= 7

> A small cocktail combining DPO datasets built by Argilla with [distilabel](https://github.com/argilla-io/distilabel). The goal of this dataset is having a small, high-quality DPO dataset by filtering only highly rated chosen responses.

<div>
<img src="https://cdn-uploads.huggingface.co/production/uploads/60420dccc15e823a685f2b03/Csd2-zPji7iwIxyz6UFe1.webp">
</div>

<p align="center">
<a href="https://github.com/argilla-io/distilabel">
<img src="https://raw.githubusercontent.com/argilla-io/distilabel/main/docs/assets/distilabel-badge-light.png" alt="Built with Distilabel" width="200" height="32"/>
</a>
</p>

## Datasets mixed

This dataset mixes the following datasets:

* [`argilla/Capybara-Preferences`](https://huggingface.co/datasets/argilla/Capybara-Preferences): random sample of highly scored chosen responses (>=4).
* [`argilla/distilabel-intel-orca-dpo-pairs`](https://huggingface.co/datasets/argilla/distilabel-intel-orca-dpo-pairs): random sample of highly scored chosen responses (>=7).
* [`argilla/ultrafeedback-binarized-preferences-cleaned`](https://huggingface.co/datasets/argilla/ultrafeedback-binarized-preferences-cleaned): random sample of highly scored chosen responses (>=4).

The samples have been randomly selected from the original datasets with a proportion of 0.33 each, as can be seen via the `dataset` column of the dataset.

原始信息汇总

DPO Mix 21K Dataset 概述

数据集基本信息

• 语言: 英语
• 许可证: MIT
• 大小分类: 10K<n<100K
• 标签: distilabel, synthetic, dpo

数据集特征

• dataset: 字符串类型
• chosen: 列表类型，包含
  • content: 字符串类型
  • role: 字符串类型
• rejected: 列表类型，包含
  • content: 字符串类型
  • role: 字符串类型
• chosen_rating: 浮点型，32位
• rejected_rating: 浮点型，32位

数据集分割

• 训练集:
  • 大小: 120528535.0 字节
  • 样本数: 20250
• 测试集:
  • 大小: 4555903.0 字节
  • 样本数: 750

下载与数据集大小

• 下载大小: 198089631 字节
• 数据集大小: 125084438.0 字节

配置信息

• 默认配置:
  • 训练数据路径: data/train-*
  • 测试数据路径: data/test-*

数据集混合来源

• argilla/Capybara-Preferences: 随机抽样的高评分选择响应 (>=4)
• argilla/distilabel-intel-orca-dpo-pairs: 随机抽样的高评分选择响应 (>=7)
• argilla/ultrafeedback-binarized-preferences-cleaned: 随机抽样的高评分选择响应 (>=4)

每个来源的样本按0.33的比例随机选择。