444

# Qwen3 微调数据集

本仓库包含用于Qwen3模型微调的高质量数据集。这些数据集经过精心选择，适用于增强大语言模型在各种任务上的性能，特别是推理和知识密集型任务。

## 数据集列表

### 1. FineTome-100k

一个高质量的指令遵循数据集，专为大语言模型微调设计。

- **来源**: [mlabonne/FineTome-100k](https://huggingface.co/datasets/mlabonne/FineTome-100k)
- **样本数量**: 100,000
- **格式**: 包含对话格式的训练数据，每个样本包含对话内容、来源和质量分数
- **特点**: 数据质量高，覆盖广泛的指令类型和领域

### 2. OpenMathReasoning-mini

专注于数学推理能力的微调数据集，包含各种数学问题及其详细解答过程。

- **来源**: [unsloth/OpenMathReasoning-mini](https://huggingface.co/datasets/unsloth/OpenMathReasoning-mini)
- **样本数量**: 19,252
- **格式**: 包含数学问题、期望答案、问题类型、解答过程等字段
- **特点**: 增强模型的数学推理和思维链(Chain-of-Thought)能力

## 使用方法

这些数据集适用于Qwen3系列模型的监督微调(SFT)，可以通过以下方式使用：

```python
from datasets import load_from_disk

# 加载FineTome数据集
finetome_dataset = load_from_disk("FineTome")

# 加载OpenMathReasoning数据集
math_dataset = load_from_disk("OpenMathReasoning")
```

## 许可说明

请注意，使用这些数据集时需遵循原始数据集的许可条款。在商业应用前，请确认相应的使用权限。

# Qwen3 Fine-tuning Datasets

This repository contains high-quality datasets for fine-tuning the Qwen3 series models. These datasets have been carefully selected to enhance the performance of large language models across various tasks, especially reasoning and knowledge-intensive tasks.

## Dataset List

### 1. FineTome-100k
A high-quality instruction-following dataset designed specifically for fine-tuning large language models.
- **Source**: [mlabonne/FineTome-100k](https://huggingface.co/datasets/mlabonne/FineTome-100k)
- **Sample Count**: 100,000
- **Format**: Contains conversational training data, with each sample including conversation content, source, and quality score
- **Features**: High data quality, covering a wide range of instruction types and domains

### 2. OpenMathReasoning-mini
A fine-tuning dataset focused on mathematical reasoning capabilities, containing various math problems and their detailed solution processes.
- **Source**: [unsloth/OpenMathReasoning-mini](https://huggingface.co/datasets/unsloth/OpenMathReasoning-mini)
- **Sample Count**: 19,252
- **Format**: Includes fields such as mathematical problems, expected answers, problem types, and solution processes
- **Features**: Enhances the model's mathematical reasoning and Chain-of-Thought capabilities

## Usage Instructions
These datasets are suitable for supervised fine-tuning (SFT) of the Qwen3 series models, and can be used via the following approach:
python
from datasets import load_from_disk

# Load the FineTome dataset
finetome_dataset = load_from_disk("FineTome")

# Load the OpenMathReasoning dataset
math_dataset = load_from_disk("OpenMathReasoning")

## License Notice
Please note that the use of these datasets must comply with the license terms of the original datasets. Please confirm the corresponding usage rights prior to commercial application.