Gemma is Google DeepMind’s family of open-weight large language models, built from the same research and technology as Gemini. This repository contains the official gemma Python package: a JAX library for using and fine-tuning Gemma models, published on PyPI with documentation and examples. The repository has about 5,800 stars on GitHub and is actively maintained, with updates through 2026, making it the canonical starting point for developers who want to run Google’s open models in a JAX workflow.
Google releases the Gemma model family as open weights so researchers and developers can build on the same technology behind Gemini in their own environments. This library focuses on the JAX ecosystem: install JAX for CPU, GPU or TPU, install the gemma package, and you can load checkpoints, run inference and fine-tune models with the performance characteristics JAX is known for, including just-in-time compilation and easy scaling across accelerators. The documentation walks through examples and use cases, from basic generation to tuning, and the maintainers invite issues and feedback directly on GitHub. For teams standardized on JAX โ common in research labs and at Google itself โ this is the supported path to Gemma, complementing the other framework ports in the wider ecosystem.
The Gemma library is free and open source under the Apache-2.0 license, and the Gemma model weights are openly available under Google’s Gemma terms. Your only costs are the compute you run the models on.
Gemma (the library) is for JAX developers, researchers and Google-ecosystem teams who want to run, evaluate and fine-tune Google DeepMind’s open-weight language models, whether for experiments, products or model research.