Introducing Gemma: Google’s New AI Model

Google has recently launched Gemma, a new artificial intelligence model, signaling a fresh direction for AI development. Aimed at offering developers and researchers more flexibility, Gemma represents a departure from Google’s previous AI models, focusing on openness and accessibility.


What Exactly Is Google Gemma?


Gemma comprises two main models: Gemma 2B and Gemma 7B. These models are designed to be lightweight and efficient, allowing them to run on a variety of platforms, including personal laptops and Google Cloud.

Jeanine Banks, VP & GM at Developer X and DevRel, and Tris Warkentin, Director at Google DeepMind, have shared their excitement about Gemma’s potential to foster innovation and responsible AI development.


How Can Developers Use Gemma?


Gemma is open-source, meaning developers worldwide can access and use these models for various applications, from creating simple chatbots to enhancing language processing tasks. Google has also provided extensive resources, including Colab and Kaggle notebooks, to help developers get started.

“Open models are essential for the democratisation of AI technology,” says Jeanine Banks. This sentiment is echoed by developers who see Gemma as a valuable resource for creating more diverse and innovative AI applications.


Is Gemma Safe To Use?


Safety and responsibility are at the forefront of Gemma’s design. Google has implemented measures to ensure that the models exclude sensitive information and align with ethical guidelines. A Responsible Generative AI Toolkit accompanies Gemma, offering tools for safety classification, debugging, and best practice guidance.

Tris Warkentin emphasises the importance of responsible AI development, stating, “We’ve conducted extensive evaluations to minimise risks associated with open models.” These efforts reflect Google’s commitment to creating AI technologies that are safe and beneficial for society.

What Makes Gemma Different?


Gemma stands out due to its open-source nature and the flexibility it offers developers. Unlike its predecessor, Gemini, which is a closed model, Gemma can be used and modified freely, promoting a more collaborative and inclusive approach to AI development. This development by Google could greatly influence how AI technologies are developed and deployed in the future.


What Makes Gemma Compatible With Various Frameworks, Tools, And Hardware?


Google Gemma models have been crafted with versatility in mind, ensuring they work smoothly across various frameworks, tools, and hardware. Jeanine Banks, VP & GM at Developer X and DevRel, highlights, “Gemma models are designed to be easily integrated into your current projects, supporting major AI frameworks like JAX, PyTorch, and TensorFlow.”

This adaptability means developers can incorporate Gemma into their workflow without needing to switch between different tools or platforms, facilitating a seamless development experience.

On top of this, Tris Warkentin, Director at Google DeepMind, discuses the models’ compatibility with diverse hardware setups: “Whether you’re coding on a laptop or deploying on Google Cloud, Gemma’s performance remains top-notch, thanks to our optimisations across NVIDIA GPUs and Google Cloud TPUs.”

This broad hardware support, including for local RTX AI PCs, ensures that developers have the flexibility to work on projects of any scale, from personal experiments to large-scale commercial applications.