Last Updated on February 22, 2024 8:34 am by Laszlo Szabo / NowadAIs | Published on February 22, 2024 by Laszlo Szabo / NowadAIs
Google Gemma: New Series of Open Models for AI Research Run Locally – Key Notes
- Google announces Gemma, a new series of open models for AI research.
- Gemma models are advanced, lightweight, and built on Google DeepMind’s technology.
- Accompanied by a Responsible Generative AI Toolkit for addressing potential issues.
Overview & Technical Details of Google Gemma
Google has announced Gemma, their new open models designed for AI research.
It is a collection of lightweight and advanced models created by Google DeepMind and other teams within the company. These models are built using the same technology and research as the Gemini models and include both technical and infrastructure components.
Gemma is a new family of open models that help developers and researchers build AI. Along with the lightweight models, we’re launching tooling that encourages collaboration and a guide to responsible use of these models. Learn more → https://t.co/DCRQ62WYdv pic.twitter.com/jLTlbawDVE
— Google (@Google) February 21, 2024
The name Gemma is derived from Latin, meaning
“precious stone.”
Currently, Gemma 2B and Gemma 7B are available, along with pre-trained and instruction-tuned variations. For Gemma 2B, Google has removed any personal or sensitive data from the training sets and has undergone rigorous testing to ensure responsible behavior.
More details about these evaluations can be found online.
Compatibility, Optimization and Support
The models have also been fine-tuned and reinforced with human feedback to ensure responsible behavior. Gemma can be used on laptops, desktops, IoT devices, mobile devices, and in the cloud.
It is compatible with easy deployment on Vertex AI and Google Kubernetes Engine (GKE), and has been optimized for Nvidia GPUs in addition to Google Cloud TPUs.
According to Google, Gemma offers top-notch performance compared to other open models and even outperforms larger models on important benchmarks. A technical report with more information on Gemma’s performance, dataset composition, and modeling methods is available.
Google allows responsible commercial use and distribution of Gemma for all organizations, regardless of size. They have also released a Responsible Generative AI Toolkit with guidance, a model debugging tool, and safety classification to help address any potential issues.
Gemma is now available worldwide, and Google plans to expand the model family in the future.
Definitions
- Google Gemma: It’s a new series of open AI research models created by Google DeepMind, designed to be lightweight, advanced, and responsible.
- Google Kubernetes Engine (GKE): A managed environment for running containerized applications using Google’s infrastructure, simplifying the deployment, scaling, and operations of such applications.
- Vertex AI: A managed machine learning platform by Google that allows developers to deploy and maintain AI models more efficiently.
- Nvidia GPU: High-performance graphics processing units designed by Nvidia, widely used for computing tasks in AI and deep learning.
- Google Cloud TPU: Custom-built hardware designed by Google to accelerate machine learning workloads, offering high throughput for training and inference tasks.
Frequently Asked Questions
- What is Google Gemma?
Google Gemma is a new open model initiative by Google for advancing AI research, offering lightweight, advanced models for various applications. - How does Google Gemma contribute to AI research?
Google Gemma offers state-of-the-art performance, promoting responsible AI development with models that are rigorously tested for ethical use. - Can I use Google Gemma for commercial purposes?
Yes, Google allows the responsible commercial use of Gemma, aiming to make advanced AI research accessible to all organizations. - What makes Google Gemma unique?
Gemma stands out for its performance, responsible AI focus, compatibility across devices and platforms, and optimization for both Nvidia GPUs and Google Cloud TPUs. - How can I ensure responsible use of Google Gemma?
Google provides a Responsible Generative AI Toolkit with Gemma, including guidance, a debugging tool, and safety classification to address potential issues.
