Easy Dataset is an open-source application for creating the datasets used to fine-tune large language models, power retrieval-augmented generation, and evaluate model quality. Instead of hand-writing training examples, you feed it your domain documents and it turns them into structured, high-quality data through an intuitive visual interface. The project is actively developed on GitHub, where it has gathered more than 15,000 stars, with updates continuing into 2026.
Building fine-tuning data is usually the slowest part of adapting an LLM to a specialty domain. Easy Dataset automates the pipeline: it parses documents in formats such as PDF, Markdown, DOCX, TXT, and EPUB, splits them intelligently, generates question-answer pairs and other training structures with the help of an LLM, and provides cleaning and augmentation tools to raise data quality. Version 1.7 added a full evaluation workbench that converts domain documents into test sets and runs multi-dimensional evaluations, including human blind tests, so you can measure a fine-tuned model or a RAG pipeline’s recall before shipping it. Everything runs locally as a desktop-style web app, and finished datasets export in the formats popular training frameworks expect.
Easy Dataset is free and open source. You run it yourself and only pay for any LLM API calls you choose to make during data generation, which keeps it inexpensive for individuals, students, and small teams experimenting with fine-tuning.
Easy Dataset is aimed at ML engineers, researchers, and domain specialists who need custom training or evaluation data without building a data pipeline from scratch. If you are fine-tuning a model on company documents, textbooks, manuals, or research papers, it removes most of the tedious preparation work.