llamafactory/DPO-En-Zh-20k
收藏Hugging Face2024-06-07 更新2025-04-12 收录
下载链接:
https://hf-mirror.com/datasets/llamafactory/DPO-En-Zh-20k
下载链接
链接失效反馈官方服务:
资源简介:
---
license: apache-2.0
task_categories:
- text-generation
language:
- en
- zh
tags:
- dpo
- orpo
- rlhf
- llama-factory
size_categories:
- 10K<n<100K
configs:
- config_name: en
data_files:
- split: train
path: dpo_en.json
- config_name: zh
data_files:
- split: train
path: dpo_zh.json
dataset_info:
- config_name: en
features:
- name: conversations
list:
- name: from
dtype: string
- name: value
dtype: string
- name: chosen
struct:
- name: from
dtype: string
- name: value
dtype: string
- name: rejected
struct:
- name: from
dtype: string
- name: value
dtype: string
- config_name: zh
features:
- name: conversations
list:
- name: from
dtype: string
- name: value
dtype: string
- name: chosen
struct:
- name: from
dtype: string
- name: value
dtype: string
- name: rejected
struct:
- name: from
dtype: string
- name: value
dtype: string
---
This dataset is composed by
- 4,000 examples of [argilla/distilabel-capybara-dpo-7k-binarized](https://huggingface.co/datasets/argilla/distilabel-capybara-dpo-7k-binarized) with chosen score>=4.
- 3,000 examples of [argilla/distilabel-intel-orca-dpo-pairs](https://huggingface.co/datasets/argilla/distilabel-intel-orca-dpo-pairs) with chosen score>=8.
- 3,000 examples of [argilla/ultrafeedback-binarized-preferences-cleaned](https://huggingface.co/datasets/argilla/ultrafeedback-binarized-preferences-cleaned) with chosen score>=4.
- 10,000 examples of [wenbopan/Chinese-dpo-pairs](https://huggingface.co/datasets/wenbopan/Chinese-dpo-pairs).
You can use it in [LLaMA Factory](https://github.com/hiyouga/LLaMA-Factory) by specifying `dataset: dpo_mix_en,dpo_mix_zh`.
许可证:Apache-2.0
任务类别:
- 文本生成
语言:
- 英语
- 汉语
标签:
- 直接偏好优化(Direct Preference Optimization,DPO)
- 比值比偏好优化(Odds Ratio Preference Optimization,ORPO)
- 基于人类反馈的强化学习(Reinforcement Learning from Human Feedback,RLHF)
- LLaMA Factory
样本规模区间:1万 < 样本量 < 10万
配置项:
- 配置名称:en
数据文件:
- 拆分集:训练集
路径:dpo_en.json
- 配置名称:zh
数据文件:
- 拆分集:训练集
路径:dpo_zh.json
数据集信息:
- 配置名称:en
特征字段:
- 字段名:conversations,为列表类型,每个元素包含:
- 字段:from,数据类型:字符串
- 字段:value,数据类型:字符串
- 字段名:chosen,为结构体类型,包含:
- 字段:from,数据类型:字符串
- 字段:value,数据类型:字符串
- 字段名:rejected,为结构体类型,包含:
- 字段:from,数据类型:字符串
- 字段:value,数据类型:字符串
- 配置名称:zh
特征字段:
- 字段名:conversations,为列表类型,每个元素包含:
- 字段:from,数据类型:字符串
- 字段:value,数据类型:字符串
- 字段名:chosen,为结构体类型,包含:
- 字段:from,数据类型:字符串
- 字段:value,数据类型:字符串
- 字段名:rejected,为结构体类型,包含:
- 字段:from,数据类型:字符串
- 字段:value,数据类型:字符串
本数据集由以下部分构成:
- 4000条样本来自数据集[argilla/distilabel-capybara-dpo-7k-binarized](https://huggingface.co/datasets/argilla/distilabel-capybara-dpo-7k-binarized),且优选回复得分≥4。
- 3000条样本来自数据集[argilla/distilabel-intel-orca-dpo-pairs](https://huggingface.co/datasets/argilla/distilabel-intel-orca-dpo-pairs),且优选回复得分≥8。
- 3000条样本来自数据集[argilla/ultrafeedback-binarized-preferences-cleaned](https://huggingface.co/datasets/argilla/ultrafeedback-binarized-preferences-cleaned),且优选回复得分≥4。
- 10000条样本来自数据集[wenbopan/Chinese-dpo-pairs](https://huggingface.co/datasets/wenbopan/Chinese-dpo-pairs)。
你可通过指定`dataset: dpo_mix_en,dpo_mix_zh`在[LLaMA Factory](https://github.com/hiyouga/LLaMA-Factory)中使用本数据集。
提供机构:
llamafactory


