Google announced the launch of Gemma, a new open source AI model. Developed using the technology that underpinned Google's Gemini AI models, Gemma aims to provide developers with advanced tools to create AI applications conscientiously.
Available in two configurations, Gemma 2B and Gemma 7B, this open source AI model offers pre-trained and instruction-tuned variants to operate efficiently on standard computing devices such as laptops and desktops. Google highlights that despite their compact size, both Gemma models demonstrate superior performance over larger counterparts and outshine other open models.
In conjunction with Gemma, Google has also introduced the Responsible Generative AI Toolkit. This comprehensive package includes a debugging tool and a guideline booklet outlining best practices for AI development, drawing from Google's extensive experience in the field. This initiative underscores Google's commitment to ensuring that AI technologies are developed with safety and ethical considerations at the forefront.
To ensure the highest standards of safety, Google employed automated techniques to remove personal information from the training data of the Gemma models. Moreover, reinforcement learning, guided by human feedback, was utilized to refine the instruction-tuned variants of Gemma, guaranteeing they adhere to responsible behavioral patterns.
Google has made Gemma accessible through several platforms, including Colab and Kaggle notebooks, alongside integration with popular tools such as Hugging Face, MaxText, NVIDIA NeMo and TensorRT-LLM.
The introduction of Gemma has set new benchmarks, outperforming all other models in its size categories. This achievement has ignited the AI community with a renewed sense of enthusiasm. Clem Delangue, co-founder & CEO of Hugging Face shared in his X account:
Google is back baby! Taking the first spot for open models on the Hugging Face LLM leaderboard for its sizes (2B & 7B).
In a groundbreaking collaboration, NVIDIA has joined forces with Google to optimize Gemma across NVIDIA's AI platforms, significantly enhancing Gemma's performance on NVIDIA GPUs. This collaboration extends Gemma's reach, allowing developers to leverage the model on over 100 million NVIDIA RTX GPUs globally, including in cloud environments and on local workstations.
NVIDIA's support for Gemma extends to the cloud, with Google Cloud's A3 instances and the forthcoming deployment of NVIDIA’s H200 Tensor Core GPUs, promising unparalleled memory and speed. Developers stand to benefit from NVIDIA's comprehensive ecosystem of tools for fine-tuning and deploying Gemma in production applications, marking a significant step forward in accessible, high-performance AI development.
Additionally, NVIDIA plans to integrate Gemma into its Chat with RTX technology, offering users enhanced generative AI capabilities on RTX-powered PCs. This integration will allow for personalized chatbots that can operate with data stored locally on a user's PC, ensuring privacy and rapid response times without the need for cloud-based services.