Responsible by Design
Welcome to the world of Gemma, where responsible AI design is at the forefront. At Gemma, we adhere to our AI Principles to ensure that our pre-trained models are safe and reliable. We employ cutting-edge automated techniques to filter out sensitive data and personal information from training sets. Additionally, our models undergo extensive fine-tuning and reinforcement learning from human feedback to align with responsible behaviors. To mitigate risks, we conduct thorough evaluations, including manual red-teaming and automated adversarial testing, outlined in our Model Card.
Exciting news! Along with Gemma, we are launching the Responsible Generative AI Toolkit to assist developers and researchers in building safe and responsible AI applications. The toolkit includes:
- Safety classification: Explore a novel methodology for building robust safety classifiers with minimal examples.
- Debugging: Use our model debugging tool to investigate Gemma’s behavior and address potential issues.
- Guidance: Access best practices for model builders based on Google’s vast experience.
Optimized for Performance
Gemma models are designed for versatility and optimized performance across various frameworks, tools, and hardware. You can fine-tune Gemma models to suit your specific application needs, such as summarization or retrieval-augmented generation (RAG). Gemma supports:
- Multi-framework tools: Utilize reference implementations for inference and fine-tuning across Keras 3.0, PyTorch, JAX, and Hugging Face Transformers.
- Cross-device compatibility: Run Gemma models on popular devices, from laptops to cloud infrastructure, ensuring accessibility.
- Cutting-edge hardware platforms: Partnered with NVIDIA to optimize Gemma for NVIDIA GPUs for unrivaled performance.
- Optimized for Google Cloud: Vertex AI provides a wide MLOps toolset with deployment options across GPU, TPU, and CPU infrastructure.
Free Credits for Innovation
We believe in fostering an inclusive community of developers and researchers. Begin your journey with Gemma today with free access on Kaggle, Colab notebooks, and $300 credits for first-time Google Cloud users. Research projects can also apply for Google Cloud credits to accelerate their work.
Getting Started
Ready to explore Gemma? Access quickstart guides and more on ai.google.dev/gemma.
Stay tuned for upcoming events and opportunities with Gemma as we continue to expand our model family. The possibilities are endless!
Create, innovate, and build responsibly with Gemma!