Gemma is a series of lightweight open AI models developed by Google DeepMind and other teams, designed to assist developers and researchers in building responsible AI app...
GemmaIn today’s rapidly evolving AI landscape, Google once again leads the way with the introduction of its new lightweight open AI model series—Gemma. This series aims to provide developers and researchers with more efficient and responsible AI tools, empowering them to achieve innovative breakthroughs in their respective fields.
Gemma models are designed to be lightweight, facilitating operation across various computing environments, including personal computers and workstations.
The weights of Gemma models are open, allowing users to use and distribute them commercially under the terms of the license agreement.
Offers both pre-trained models and instruction-tuned versions, the latter ensuring responsible model behavior through Reinforcement Learning from Human Feedback (RLHF).
Gemma supports major AI frameworks such as JAX, PyTorch, and TensorFlow, providing toolchains through Keras 3.0 to simplify inference and Supervised Fine-Tuning (SFT) processes.
Designed in accordance with Google’s AI principles, Gemma employs automated techniques to filter sensitive information from training data and undergoes a series of security assessments, including red teaming and adversarial testing.
Gemma models are optimized for hardware platforms like NVIDIA GPUs and Google Cloud TPUs, ensuring high performance across different devices.
Google offers free resources on platforms like Kaggle and Colab, as well as Google Cloud credits, encouraging developers and researchers to innovate and conduct research using Gemma.
Gemma models can operate on various devices, including laptops, desktops, IoT devices, and the cloud, supporting a wide range of AI functionalities.
Google has also released the Responsible Generative AI Toolkit to help developers build safe and responsible AI applications, including safety classifiers, debugging tools, and application guidelines.
