sumuks/helpsteer3-dpo-style
收藏Hugging Face2026-03-26 更新2026-03-29 收录
下载链接:
https://hf-mirror.com/datasets/sumuks/helpsteer3-dpo-style
下载链接
链接失效反馈官方服务:
资源简介:
---
pretty_name: HelpSteer3 DPO Style
language:
- en
- zh
- ko
- fr
- es
- ru
- ja
- de
- it
- pt
- pl
- id
- nl
- vi
license: cc-by-4.0
task_categories:
- text-generation
tags:
- dpo
- preference-optimization
- human-feedback
- reward-modeling
- helpsteer
- nvidia
size_categories:
- 10K<n<100K
configs:
- config_name: default
data_files:
- split: train
path: train-00000-of-00001.parquet
- split: test
path: test-00000-of-00001.parquet
---
# Dataset Card for HelpSteer3 DPO Style
## Dataset Summary
This dataset is derived from `nvidia/HelpSteer3` using the `preference` config.
Each source row contains a chat-style `context`, two candidate responses, and a signed `overall_preference` score.
This conversion keeps only strong preferences with `abs(overall_preference) >= 2` and maps them into DPO-style `chosen` and `rejected` rows.
The original HelpSteer3 `validation` split is written as `test` here to match the train/test convention used elsewhere in this repo.
## Dataset Structure
- Train source rows: 38459
- Test source rows: 2017
- Train DPO rows: 23959
- Test DPO rows: 1288
- Total DPO rows: 25247
- Dropped weak-preference rows: 15229
Each row contains these key fields:
- `prompt`: Rendered conversation transcript from the source `context`.
- `chosen`: Preferred assistant response chosen from `response1` or `response2`.
- `rejected`: Less-preferred assistant response chosen from the other source response.
- `difficulty`: `1 / abs(overall_preference)`, so `0.5` for `±2` and `0.333...` for `±3`.
## Construction Notes
- Negative `overall_preference` values mean `response1` is preferred.
- Positive `overall_preference` values mean `response2` is preferred.
- Rows with scores `-1`, `0`, and `1` are dropped as too weak for this dataset.
- Difficulty uses the absolute preference value because the sign only indicates which side won, not how hard the pair is.
- No prompt-level regrouping is needed because HelpSteer3 already ships train and validation splits.
---
pretty_name: HelpSteer3 DPO风格
language:
- 英语
- 中文
- 韩语
- 法语
- 西班牙语
- 俄语
- 日语
- 德语
- 意大利语
- 葡萄牙语
- 波兰语
- 印尼语
- 荷兰语
- 越南语
license: 知识共享署名4.0(CC BY 4.0)
task_categories:
- 文本生成
tags:
- DPO(Direct Preference Optimization)
- 偏好优化
- 人类反馈
- 奖励建模
- HelpSteer3
- 英伟达(NVIDIA)
size_categories:
- 10000 < n < 100000
configs:
- config_name: 默认
data_files:
- split: 训练集
path: train-00000-of-00001.parquet
- split: 测试集
path: test-00000-of-00001.parquet
---
# HelpSteer3 DPO风格数据集卡片
## 数据集概述
本数据集源自`英伟达(NVIDIA)/HelpSteer3`,通过`preference`配置生成。每条原始源数据行均包含对话式`上下文(context)`、两条候选回复以及带符号的`整体偏好评分(overall_preference)`。本次转换仅保留`整体偏好评分(overall_preference)绝对值≥2`的强偏好样本,并将其映射为符合DPO(Direct Preference Optimization)格式的`选中(chosen)`与`拒选(rejected)`数据行。
原HelpSteer3的`验证集(validation)`拆分在本数据集中被命名为`测试集(test)`,以匹配本仓库其他模块采用的训练/测试拆分惯例。
## 数据集结构
- 训练源数据行:38459条
- 测试源数据行:2017条
- 训练DPO格式数据行:23959条
- 测试DPO格式数据行:1288条
- DPO格式总数据行:25247条
- 过滤掉的弱偏好数据行:15229条
每条数据行包含以下关键字段:
- `prompt`:源自原始`上下文`的格式化对话文本。
- `chosen`:从`response1`或`response2`中选出的最优助手回复。
- `rejected`:来自另一候选回复的次优助手回复。
- `difficulty`:计算公式为`1 / abs(整体偏好评分(overall_preference))`,例如评分为±2时难度值为0.5,评分为±3时难度值为0.333...。
## 构建说明
- 当`整体偏好评分(overall_preference)`为负值时,表示`response1`为更优回复。
- 当`整体偏好评分(overall_preference)`为正值时,表示`response2`为更优回复。
- 评分为-1、0或1的样本因偏好强度过弱,已被过滤出本数据集。
- 难度计算采用偏好评分的绝对值,因为评分符号仅用于指示哪条回复更优,而非衡量该样本的难度。
- 无需对`prompt`进行分组重构,因为原HelpSteer3数据集已自带训练与验证拆分。
提供机构:
sumuks


