stack-exchange-paired

# StackExchange Paired

This is a processed version of the [`HuggingFaceH4/stack-exchange-preferences`](https://huggingface.co/datasets/HuggingFaceH4/stack-exchange-preferences). The following steps were applied:

- Parse HTML to Markdown with `markdownify`
- Create pairs `(response_j, response_k)` where j was rated better than k
- Sample at most 10 pairs per question
- Shuffle the dataset globally

## 示例代码
```python
from modelscope import MsDataset
from modelscope.utils.constant import DownloadMode

ds = MsDataset.load('AI-ModelScope/stack-exchange-paired',subset_name='default', split='train', download_mode=DownloadMode.FORCE_REDOWNLOAD)
print(next(iter(ds)))
```

This dataset is designed to be used for preference learning. The processing notebook is in [the repository](https://huggingface.co/datasets/lvwerra/stack-exchange-paired/tree/main) as well.

# StackExchange 配对数据集（StackExchange Paired）

本数据集是对 [`HuggingFaceH4/stack-exchange-preferences`](https://huggingface.co/datasets/HuggingFaceH4/stack-exchange-preferences) 数据集的加工版本。本次加工采用了以下步骤：
- 使用`markdownify`工具将HTML格式转换为Markdown格式
- 构建配对样本`(response_j, response_k)`，其中response_j的评分高于response_k
- 每个问题最多采样10组配对样本
- 对全数据集进行全局洗牌

## 示例代码
python
from modelscope import MsDataset
from modelscope.utils.constant import DownloadMode

ds = MsDataset.load('AI-ModelScope/stack-exchange-paired',subset_name='default', split='train', download_mode=DownloadMode.FORCE_REDOWNLOAD)
print(next(iter(ds)))

本数据集专为偏好学习（preference learning）任务设计，其加工Jupyter笔记本同样可在[该仓库](https://huggingface.co/datasets/lvwerra/stack-exchange-paired/tree/main)中获取。

搜集汇总

背景与挑战

背景概述

该数据集是'HuggingFaceH4/stack-exchange-preferences'的加工版本，通过解析HTML为Markdown、创建偏好响应对并采样打乱而成。它专为偏好学习任务设计，用于训练模型比较和选择更优回答。