Google introduced Gemma 2 27B, a cutting-edge family of lightweight open models, earlier this year. Built on the same research and technology used to develop the highly successful Gemini models, the Gemma family has continued to expand with the addition of CodeGemma, RecurrentGemma, and PaliGemma. These models offer unique capabilities for various AI tasks and are easily accessible through integrations with partners like Hugging Face, NVIDIA, and Ollama.
Now, Google is proud to announce the official release of Gemma 2 to researchers and developers worldwide. Available in both 9 billion (9B) and 27 billion (27B) parameter sizes, Gemma 2 boasts higher performance and greater efficiency in inference compared to the previous generation. With significant safety enhancements built in, Gemma 2 sets a new standard for open models in terms of efficiency and performance. The 27B model, in particular, offers competitive alternatives to models twice its size, delivering exceptional performance that was previously only achievable with proprietary models. Best of all, this level of performance can now be achieved on a single Google Cloud TPU host or NVIDIA H100 Tensor Core GPU, significantly reducing deployment costs.
Key Features of Gemma 2:
- Outsized performance: Gemma 2 at 27B delivers unmatched performance for its size class and even competes with models twice its size. The 9B model also outperforms other open models in its category, such as Llama 3 8B.
- Unmatched efficiency and cost savings: The 27B model of Gemma 2 runs inference efficiently at full precision on a single Google Cloud TPU host or NVIDIA GPU, reducing costs while maintaining high performance.
- Blazing fast inference across hardware: Gemma 2 is optimized to run smoothly across a range of hardware, from gaming laptops to cloud-based setups, providing accessible and budget-friendly AI deployments.
Gemma 2 is designed to easily integrate into developers’ workflows, offering a commercially-friendly license that allows for sharing and commercializing innovations. With broad framework compatibility, Gemma 2 can be used with major AI frameworks like Hugging Face Transformers, JAX, PyTorch, and TensorFlow. Deployment of Gemma 2 on Google Cloud’s Vertex AI is set to become effortless for customers starting next month.
In conclusion, Gemma 2 represents a significant advancement in AI technology, offering researchers and developers powerful tools to create innovative AI solutions responsibly. The release of Gemma 2 opens up exciting possibilities for cutting-edge projects and paves the way for future advancements in the field of artificial intelligence. Explore the new Gemma Cookbook for practical examples and guidance on utilizing Gemma 2 models for specific tasks, and discover the potential of Google’s latest open model for AI development.