ConardLi/easy-dataset

A powerful tool for creating fine-tuning datasets for Large Language Models

简体中文 | English | Türkçe

Features • Quick Start • Documentation • Contributing • License

If you like this project, please give it a Star⭐️, or buy the author a coffee => Donate ❤️!

Overview

Easy Dataset is an application specifically designed for building large language model (LLM) datasets. It features an intuitive interface, along with built-in powerful document parsing tools, intelligent segmentation algorithms, data cleaning and augmentation capabilities. The application can convert domain-specific documents in various formats into high-quality structured datasets, which are applicable to scenarios such as model fine-tuning, retrieval-augmented generation (RAG), and model performance evaluation.

News

🎉🎉 Easy Dataset Version 1.7.0 launches brand-new evaluation capabilities! You can effortlessly convert domain-specific documents into evaluation datasets (test sets) and automatically run multi-dimensional evaluation tasks. Additionally, it comes with a human blind test system, enabling you to easily meet needs such as vertical domain model evaluation, post-fine-tuning model performance assessment, and RAG recall rate evaluation. Tutorial: https://www.bilibili.com/video/BV1CRrVB7Eb4/

Features

📄 Document Processing & Data Generation

• Intelligent Document Processing: Supports PDF, Markdown, DOCX, TXT, EPUB and more formats with intelligent recognition
• Intelligent Text Splitting: Multiple splitting algorithms (Markdown structure, recursive separators, fixed length, code-aware chunking), with customizable visual segmentation
• Intelligent Question Generation: Auto-extract relevant questions from text segments, with question templates and batch generation
• Domain Label Tree: Intelligently builds global domain label trees based on document structure, with auto-tagging capabilities
• Answer Generation: Uses LLM API to generate comprehensive answers and Chain of Thought (COT), with AI optimization
• Data Cleaning: Intelligent text cleaning to remove noise and improve data quality

🔄 Multiple Dataset Types

• Single-Turn QA Datasets: Standard question-answer pairs for basic fine-tuning
• Multi-Turn Dialogue Datasets: Customizable roles and scenarios for conversational format
• Image QA Datasets: Generate visual QA data from images, with multiple import methods (directory, PDF, ZIP)
• Data Distillation: Generate label trees and questions directly from domain topics without uploading documents

📊 Model Evaluation System

• Evaluation Datasets: Generate true/false, single-choice, multiple-choice, short-answer, and open-ended questions
• Automated Model Evaluation: Use Judge Model to automatically evaluate model answer quality with customizable scoring rules
• Human Blind Test (Arena): Double-blind comparison of two models' answers for unbiased evaluation
• AI Quality Assessment: Automatic quality scoring and filtering of generated datasets

🛠️ Advanced Features

• Custom Prompts: Project-level customization of all prompt templates (question generation, answer generation, data cleaning, etc.)
• GA Pair Generation: Genre-Audience pair generation to enrich data diversity
• Task Management Center: Background batch task processing with monitoring and interruption support
• Resource Monitoring Dashboard: Token consumption statistics, API call tracking, model performance analysis
• Model Testing Playground: Compare up to 3 models simultaneously

📤 Export & Integration

• Multiple Export Formats: Alpaca, ShareGPT, Multilingual-Thinking formats with JSON/JSONL file types
• Balanced Export: Configure export counts per tag for dataset balancing
• LLaMA Factory Integration: One-click LLaMA Factory configuration file generation
• Hugging Face Upload: Direct upload datasets to Hugging Face Hub

🤖 Model Support

• Wide Model Compatibility: Compatible with all LLM APIs that follow the OpenAI format
• Multi-Provider Support: OpenAI, MiniMax, Ollama (local models), Zhipu AI, Alibaba Bailian, OpenRouter, and more
• Vision Models: Support Gemini, Claude, etc. for PDF parsing and image QA

🌐 User Experience

• User-Friendly Interface: Modern, intuitive UI designed for both technical and non-technical users
• Multi-Language Support: Complete Chinese, English, Turkish and Portuguese language support 🇹🇷
• Dataset Square: Discover and explore public dataset resources
• Desktop Clients: Available for Windows, macOS, and Linux

Quick Demo

https://github.com/user-attachments/assets/6ddb1225-3d1b-4695-90cd-aa4cb01376a8

Local Run

Download Client

Windows	MacOS	Linux
Setup.exe	Intel	M	AppImage

Install with NPM

1. Clone the repository:

   git clone https://github.com/ConardLi/easy-dataset.git
   cd easy-dataset

2. Install dependencies:

   npm install

3. Start the development server:

   npm run build
npm run start

4. Open your browser and visit http://localhost:1717

Using the Official Docker Image

1. Clone the repository:

git clone https://github.com/ConardLi/easy-dataset.git
cd easy-dataset

2. Modify the docker-compose.yml file:

services:
  easy-dataset:
    image: ghcr.io/conardli/easy-dataset
    container_name: easy-dataset
    ports:
'1717:1717'
    volumes:
./local-db:/app/local-db
./prisma:/app/prisma
    restart: unless-stopped

Note: It is recommended to use the local-db and prisma folders in the current code repository directory as mount paths to maintain consistency with the database paths when starting via NPM.

Note: The database file will be automatically initialized on first startup, no need to manually run npm run db:push.

3. Start with docker-compose:

docker-compose up -d

4. Open a browser and visit http://localhost:1717

Building with a Local Dockerfile

If you want to build the image yourself, use the Dockerfile in the project root directory:

1. Clone the repository:

git clone https://github.com/ConardLi/easy-dataset.git
cd easy-dataset

2. Build the Docker image:

docker build -t easy-dataset .

3. Run the container:

docker run -d \
  -p 1717:1717 \
  -v ./local-db:/app/local-db \
  -v ./prisma:/app/prisma \
  --name easy-dataset \
  easy-dataset

Note: It is recommended to use the local-db and prisma folders in the current code repository directory as mount paths to maintain consistency with the database paths when starting via NPM.

Note: The database file will be automatically initialized on first startup, no need to manually run npm run db:push.

4. Open a browser and visit http://localhost:1717

Documentation

• View the demo video of this project: Easy Dataset Demo Video
• For detailed documentation on all features and APIs, visit our Documentation Site
• View the paper of this project: Easy Dataset: A Unified and Extensible Framework for Synthesizing LLM Fine-Tuning Data from Unstructured Documents

Community Practice

• Complete test set generation and model evaluation with Easy Dataset
• Easy Dataset × LLaMA Factory: Enabling LLMs to Efficiently Learn Domain Knowledge
• Easy Dataset Practical Guide: How to Build High-Quality Datasets?
• Interpretation of Key Feature Updates in Easy Dataset
• Foundation Models Fine-tuning Datasets: Basic Knowledge Popularization

Contributing

We welcome contributions from the community! If you'd like to contribute to Easy Dataset, please follow these steps:

1. Fork the repository 2. Create a new branch (git checkout -b feature/amazing-feature) 3. Make your changes 4. Commit your changes (git commit -m 'Add some amazing feature') 5. Push to the branch (git push origin feature/amazing-feature) 6. Open a Pull Request (submit to the DEV branch)

Please ensure that tests are appropriately updated and adhere to the existing coding style.

Join Discussion Group & Contact the Author

https://docs.easy-dataset.com/geng-duo/lian-xi-wo-men

License

This project is licensed under the AGPL 3.0 License - see the LICENSE file for details.

Citation

If this work is helpful, please kindly cite as:

@misc{miao2025easydataset,
  title={Easy Dataset: A Unified and Extensible Framework for Synthesizing LLM Fine-Tuning Data from Unstructured Documents},
  author={Ziyang Miao and Qiyu Sun and Jingyuan Wang and Yuchen Gong and Yaowei Zheng and Shiqi Li and Richong Zhang},
  year={2025},
  eprint={2507.04009},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2507.04009}
}

Star History