Google is taking a further step towards AI accessibility with the presentation of Gemma on Wednesday, a new family of open models promising exceptional performance. All in collaboration with another giant, NVIDIA.
Just a few days after the surprise launch of Gemini 1.5 and 1.5 Pro, Google is (already) launching a new family of open models in Vertex AI, the service that allows you to use the Mountain View firm’s generative artificial intelligence. It is called “Gemma” and offers cutting-edge performance, with a 2B model (two billion parameters) and a 7B model (7 billion).
Google offers more than 130 AI models!
This Wednesday, February 21, 2024, Google unveiled Gemma, a family of open models integrated into Vertex AI, which shares the technology of the Gemini models. Gemma offers two sizes, 2B and 7B, with pre-trained variants for research and development. Google’s Vertex AI platform, which can also be tried for free, now offers more than 130 models.
In its idea of making AI more accessible, Google is establishing, with Gemma 2B and Gemma 7B, new performance standards among open models. Concretely, it is possible to create Gemma models from Vertex AI, then run them on Google Kubernetes Engine (GKE), the American giant’s containerized application deployment environment.
From Vertex AI, developers will be able to optimize Gemma models for specific use cases. They will be able to better reduce their operational costs, Google promises, and offer an improved user experience for real-time generative AI applications.
A collaboration with NVIDIA to enhance Gemma’s performance
Gemma templates have been designed to be compatible with a whole host of Google Cloud developer favorites, including Hugging Face Transformers, Kaggle, Colab, PyTorch, JAX and Keras 3.0. This compatibility is also supposed to allow Google models to work on different devices, from a laptop to a workstation, including Google Cloud.
Google collaborated with NVIDIA to maximize the performance of Gemma, optimized for NVIDIA GPUs. The chips and graphics card juggernaut also announced that it had launched optimizations on all of its AI platforms. “ Google and NVIDIA teams worked closely to accelerate the performance of Gemma, Google’s revolutionary new open language model “, explains NVIDIA.
Developers can now build generative AI applications, deploy custom models in portable containers, and quickly integrate AI infrastructure to meet the most stringent training and inference requirements. We can no longer stop Google, nor progress.
Download
- A powerful generation model
- A knowledge base updated in real time
- Free and integrated into the Google ecosystem
Google Gemini is an AI chatbot intrinsically connected to the web from which it draws the majority of its knowledge. The main advantage of the service is that it is completely free and offers image recognition. The gradual integration into the Google ecosystem should make it a most capable chatbot for a variety of tasks.
Google Gemini is an AI chatbot intrinsically connected to the web from which it draws the majority of its knowledge. The main advantage of the service is that it is completely free and offers image recognition. The gradual integration into the Google ecosystem should make it a most capable chatbot for a variety of tasks.
Source : Google
4