WizardLM is the official repository for the Wizard family of open language models, including WizardLM for instruction following, WizardCoder for programming and WizardMath for mathematical reasoning. The GitHub project has drawn more than 9,400 stars and the model line remains a landmark in open-model research.
The Wizard models are built with Evol-Instruct, a method that automatically rewrites simple instructions into progressively more complex and diverse ones, creating high-quality training data at scale. Fine-tuning strong base models on this evolved data produced assistants that, on release, rivaled much larger proprietary systems on instruction, coding and math benchmarks. The repository collects the papers, training code, data recipes and links to released checkpoints, and it documents the whole family tree, from the original 7B models up to large mixture-of-experts releases.
The WizardLM repository and its released models are free for research and community use, distributed openly through GitHub and Hugging Face. There is no paid tier, subscription or hosted service attached to the project.
WizardLM is for researchers studying instruction tuning and data evolution, and for developers who want proven open checkpoints for chat, coding or math tasks. It rewards anyone who wants to understand how careful data generation can lift a modest base model to near-flagship quality.