Easy Dataset

// Overview

Easy Dataset is an open-source application built for creating fine-tuning datasets for large language models. It gives you an intuitive interface plus built-in document parsing, smart text splitting, and data cleaning, so you can take domain documents and turn them into high-quality structured data.

You feed it files in formats like PDF, Markdown, DOCX, TXT, or EPUB. The tool splits the text into segments, generates relevant questions, and uses an LLM API to write answers. The result is a clean dataset you can use for model fine-tuning, retrieval-augmented generation (RAG), or model evaluation.

It works with any LLM API that follows the OpenAI format, including OpenAI, Ollama for local models, Zhipu AI, OpenRouter, and others. Finished datasets export in common formats such as Alpaca and ShareGPT, and can be sent straight to a LLaMA Factory config or uploaded to the Hugging Face Hub.

// What it does

• Smart document processing for PDF, Markdown, DOCX, TXT, and EPUB, with several text-splitting algorithms (Markdown structure, recursive separators, fixed length, code-aware)
• Automatic question generation from text segments, plus an LLM-powered answer generation step that can include Chain of Thought reasoning
• Multiple dataset types: single-turn QA, multi-turn dialogue, image QA, and data distillation directly from domain topics without uploading documents
• Built-in model evaluation system with automated judge-model scoring and a human blind-test arena for comparing two models' answers
• Export to Alpaca, ShareGPT, and Multilingual-Thinking formats as JSON or JSONL, with one-click LLaMA Factory config generation and direct Hugging Face upload
• Wide model compatibility with any OpenAI-format API, including OpenAI, Ollama, Zhipu AI, Alibaba Bailian, and OpenRouter

// Getting started

You can run Easy Dataset from source with Node.js, use the official Docker image, or download a desktop client for Windows, macOS, or Linux. The steps below cover the NPM and Docker paths from the project README.

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

npm install
npm run build
npm run start

docker-compose up -d

http://localhost:1717

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Convert internal manuals, research papers, or knowledge bases into single-turn or multi-turn QA datasets for fine-tuning a domain-specific model
• Generate evaluation datasets (true/false, single-choice, short-answer, open-ended) and score model answers automatically or through a human blind test
• Prepare data for RAG pipelines by cleaning and structuring source documents into question-answer pairs
• Build image QA datasets from a folder of images or a PDF for training and testing vision models
• Export ready-to-use datasets to LLaMA Factory or upload them directly to the Hugging Face Hub

// How Easy Dataset compares

Easy Dataset alongside other open-source synthetic data tools AI/TLDR tracks, ranked by GitHub stars.