Google unveils Gemma: A new generation of open AI models for responsible development

Today, Google launches Gemma -- a new family of open AI models designed to help developers and researchers build responsible AI applications. Developed by Google DeepMind and other teams across Google, Gemma models are built on the same technology as the Gemini models.

Gemma models are available in two sizes: Gemma 2B and Gemma 7B, each with pre-trained and instruction-tuned variants. Google is also releasing a Responsible Generative AI Toolkit to guide the creation of safer AI applications. The toolkit includes a safety classification methodology, a model debugging tool, and best practices for model builders.

The models are optimized for use across various frameworks, including JAX, PyTorch, and TensorFlow through native Keras 3.0. They are compatible with a range of devices and hardware platforms, including NVIDIA GPUs and Google Cloud TPUs, ensuring industry-leading performance. Gemma models are designed to run on laptops, workstations, or Google Cloud, with easy deployment on Vertex AI and Google Kubernetes Engine (GKE).

Google has prioritized safety and responsibility in the development of Gemma. The models have undergone extensive fine-tuning and reinforcement learning from human feedback (RLHF) to align with responsible behaviors. Automated techniques were used to filter out sensitive data from training sets, and robust evaluations were conducted to assess the risk profile of the models.

To support the open community of developers and researchers, Google is offering free access to Gemma on Kaggle, a free tier for Colab notebooks, and $300 in credits for first-time Google Cloud users. Researchers can also apply for Google Cloud credits of up to $500,000 to accelerate their projects.