Fresh off the boat after Gemini 1.5, Google launches new AI model Gemma

Google is having quite a busy few months to catch up with Microsoft in the AI race. Not too long after launching Gemini and Gemini 1.5, its most capable AI model yet, the company announced Gemma, another model that’s built from the same tech used for the Gemini models.

“Developed by Google DeepMind and other teams across Google, Gemma is inspired by Gemini, and the name reflects the Latin gemma, meaning “precious stone,” Google says in the official announcement, just fresh after announcing Gemini 1.5 that could be 20 times faster than GPT-4.

Google’s Gemma AI comes in two models: Gemma 2B and Gemma 7B. These differ in size and capabilities. Both come in two flavors: “pre-trained” for general use and “instruction-tuned” for specific tasks.

But what exactly is the difference? Gemma isn’t just any open-source AI model. It’s a light model that’s specially designed to help developers and researchers build responsible AI, guided by Google’s AI Principles in mind.

You can also build safer and more ethical AI applications using Google’s Responsible Generative AI Toolkit, which also comes within this model. 

The toolkit also provides toolchains for inference and fine-tuning across popular frameworks like JAX, PyTorch, and TensorFlow, along with ready-to-use notebooks and integrations with established tools like Hugging Face and NVIDIA NeMo.