According to a new press release, Google has unveiled Gemma, a new generation of state-of-the-art open models designed for responsible AI development. Gemma, inspired by the Latin word “gemma,” meaning “precious stone,” is a family of lightweight models developed by Google DeepMind and other teams across Google. Released in two sizes, Gemma 2B and Gemma 7B, with pre-trained and instruction-tuned variants, these models can run on various platforms, from laptops to Google Cloud, ensuring broad accessibility. The release includes a Responsible Generative AI Toolkit, offering guidance and tools to foster collaboration and guide responsible use. Gemma surpasses larger models on benchmarks while adhering to rigorous safety standards.

Gemma models share technical components with Gemini, Google’s largest AI model. The models are designed with Google’s AI principles at the forefront, incorporating automated techniques to filter sensitive data and extensive evaluations, including manual red-teaming and adversarial testing. The new Responsible Generative AI Toolkit is released alongside Gemma, providing safety classification, debugging tools, and guidance for developers and researchers. Gemma supports multiple frameworks and hardware platforms, optimized for Google Cloud, and comes with free access for researchers and developers.

To encourage adoption, Google offers free access to Gemma through Kaggle and Colab, along with $300 in credits for first-time Google Cloud users. The company aims to expand the Gemma model family for diverse applications, inviting developers and researchers to connect, learn, and build with Gemma through upcoming events and opportunities.