Math-Glot-Cleaned-10K

# **Math-Glot-Cleaned-10K**

**Math-Glot-Cleaned-10K** is a curated subset of 10,000 math and code reasoning samples, extracted and cleaned from the original [NVIDIA AceReason-1.1-SFT](https://huggingface.co/datasets/NVIDIA/AceReason-1.1-SFT) dataset. This refined version focuses exclusively on high-quality, structured mathematical prompts paired with chain-of-thought style reasoning.

## Dataset Summary

* **Source**: Derived from NVIDIA's AceReason-1.1-SFT dataset
* **Total Entries**: 10,000
* **Format**: Text-to-text (input → output)
* **Modality**: Text
* **Language**: English
* **License**: Apache 2.0

Each entry in this dataset contains:

* **input**: A mathematical or computational problem, often using LaTeX-style expressions
* **output**: A detailed reasoning response, typically in chain-of-thought format

## Processing Details

The following steps were applied during dataset creation:

* **Filtered**: Retained only math- and code-related samples from the full AceReason dataset
* **Normalized**: Standardized LaTeX formatting, spacing, and markdown structure
* **Cleaned**: Removed noisy, unrelated, or improperly formatted records
* **Column Reduction**: Dropped unnecessary columns from the original dataset

## Use Cases

This dataset is intended for:

* Fine-tuning models on math reasoning tasks
* Developing instruction-following LLMs for structured problem solving
* Benchmarking reasoning capabilities of large models in math or algorithmic domains

## Example

| input | output |
| ----------------------------------------------------- | --------------------------------------------------------------------------------------------------------------- |
| Find all prime numbers $p$ such that $p^3 + 1 = k^2$. | `<think>` Okay, let's see. I need to find all prime numbers p where p cubed plus one equals a perfect square... |

## Citation

If using this dataset, please cite the original [AceReason-1.1-SFT dataset](https://huggingface.co/datasets/NVIDIA/AceReason-1.1-SFT) and credit this cleaned 10K subset for its focused structure.

# **Math-Glot-Cleaned-10K**

**Math-Glot-Cleaned-10K** 是从原始[NVIDIA AceReason-1.1-SFT](https://huggingface.co/datasets/NVIDIA/AceReason-1.1-SFT)数据集提取并清洗得到的精选子集，共包含10000条数学与代码推理样本。该优化版本仅聚焦于高质量、结构化的数学提示词与思维链（chain-of-thought）风格推理内容的配对组合。

## 数据集概述
* **来源**：衍生自NVIDIA的AceReason-1.1-SFT数据集
* **总样本数**：10,000
* **格式**：文本到文本（输入→输出）
* **模态**：文本
* **语言**：英语
* **授权协议**：Apache 2.0

该数据集的每条样本均包含：
* **输入（input）**：一道数学或计算问题，通常采用LaTeX格式的表达式
* **输出（output）**：详细的推理响应，一般采用思维链（chain-of-thought）格式

## 数据处理细节
数据集构建过程中执行了以下步骤：
* **筛选**：仅保留原始AceReason数据集中与数学和代码相关的样本
* **标准化**：统一LaTeX格式、空格排版与Markdown结构
* **清洗**：移除存在噪声、无关内容或格式错误的记录
* **列精简**：删除原始数据集中的非必要列

## 应用场景
本数据集适用于：
* 针对数学推理任务的模型微调
* 开发面向结构化问题求解的指令遵循型大语言模型（Large Language Model，LLM）
* 针对大模型在数学或算法领域的推理能力进行基准测试

## 示例
| 输入(input) | 输出(output) |
| ----------------------------------------------------- | --------------------------------------------------------------------------------------------------------------- |
| 找出所有满足 $p^3 + 1 = k^2$ 的质数 $p$ | `<think>` 好的，我们来看看。我需要找出所有质数 $p$，使得 $p$ 的三次方加1等于一个完全平方数... |

## 引用说明
若使用本数据集，请引用原始[NVIDIA AceReason-1.1-SFT数据集](https://huggingface.co/datasets/NVIDIA/AceReason-1.1-SFT)，并注明本清洗后的10K子集的结构化优化工作。