by 
Google has introduced its Gemma (on 21st February 2024), a new lineup of open AI models aimed at facilitating the development of responsible AI applications. Developed by Google, Deepmind, and other teams, Gemma models are based on the same technology as the Gemini models.
Gemma is available in two sizes, Gemma 2B and Gemma 7B, each with pre-trained and instruction-tuned variants, these models come with a Responsible Generative AI Toolkit which includes a safety classification methodology, a models debugging tool, and best practices for model builders.
It is optimized for various frameworks like JAX, PyTorch, and TensorFlow through native Keras 3.0 Gemma models that are compatible with NVIDIA GPUs and Google Cloud TPUs, ensuring top-notch performance, suitable to run on laptops, PCs, and Google Cloud.
The models have undergone extensive fine-tuning and reinforcement learning from human feedback to align with responsible behaviors. Also, there are automated techniques were employed to filter sensitive data from training sets