Google Unveils Gemma: Open AI Models Inspired by Gemini, Made Easy and Responsible
Introduction:
Developers and researchers, rejoice! Google introduces Gemma, a new family of open-source AI models designed to empower responsible AI development. Built on the foundation of Gemini, Gemma offers state-of-the-art capabilities in lightweight packages.
Key Details:
- Two model sizes: Gemma 2B and 7B, each with pre-trained and fine-tuned versions.
- Responsible tools: A toolkit guides you in building safe AI using techniques for risk assessment, debugging, and best practices.
- Easy to use: Run Gemma on laptops, workstations, or Google Cloud with pre-built Colab and Kaggle notebooks, and integrations with popular tools.
- Powerful performance: Optimized for major AI hardware like NVIDIA GPUs and Google Cloud TPUs, exceeding larger models in key benchmarks.
- Accessible to all: Free model access and credits for research and development.
Responsible by Design:
- Safe training: Google filters sensitive data and uses human feedback to refine the models.
- Robust evaluation: Red-teaming, adversarial testing, and capability assessments ensure responsible use.
- Ethical toolkit: Google provides resources to help you prioritize safety and build trustworthy AI applications.
Flexibility and Customization:
- Fine-tune with your data: Adapt Gemma to specific needs like summarization or text generation.
- Multi-framework support: Use your preferred tools like Keras, PyTorch, JAX, and Hugging Face.
- Cross-device compatibility: Run Gemma on laptops, desktops, mobile devices, and the cloud.
- Optimized hardware: Enjoy industry-leading performance on NVIDIA GPUs and Google Cloud TPUs.
Get Started:
Dive into Gemma today! Explore QuickStart guides and resources at ai.google.dev/gemma.
