Gemma, a family of responsible AI models, is introduced by Google, developed from the same technology as Gemini models. Gemma models come in sizes 2B and 7B, offering state-of-the-art performance across frameworks like JAX, PyTorch, and TensorFlow. They are available worldwide, with tools for inference and fine-tuning and integration with popular platforms. Gemma prioritizes safety and reliability, using automated techniques and robust evaluations to filter sensitive data. A new Responsible Generative AI Toolkit accompanies Gemma to help developers build safe applications. Gemma supports multi-framework tools, cross-device compatibility, and optimization for Google Cloud, offering free access and credits for researchers. Visit ai.google.dev/gemma to get started.
https://blog.google/technology/developers/gemma-open-models/