TL;DR
- Google announced it has launched an open-source AI model called Gemma.
- Gemma will come in two sizes: Gemma 2B and Gemma 7B.
- The new AI model is said to be lightweight and can outcompete significantly larger models on key benchmarks.
Despite all the recent Gemini-related news, Google isn’t slowing down on AI development for even a second. The latest from Google involves an open-source spinoff of Gemini called Gemma.
Today, Google announced it launched a family of lightweight open-source AI models built off the same technical and infrastructure components as Gemini. This family of open models is known as Gemma and comes in two sizes: Gemma 2B (2 billion parameters) and Gemma 7B (7 billion parameters).
Both models come with pre-trained and instruction-tuned variants and have been optimized to run on various AI hardware platforms, like NVIDIA GPUs and Google Cloud TPUs. Google also claims that Gemma is lightweight enough to run on developer laptops and desktops.
Despite being light, Google says the model punches above its weight. The firm boasts that Gemma not only bests other open models — like Meta’s Llama 2 — but also significantly outperforms larger models on key benchmarks.
Taking a stance similar to how it handles Gemini, the tech giant says Gemma was designed to safe and reliable. To help with that goal, it used “automated techniques to filter out certain personal information and other sensitive data from training sets.” Google states that it also used fine-tuning and reinforcement learning from human feedback to align the model with responsible behaviors and conducted evaluations to reduce risk and dangerous activities. In addition, the company is releasing a Responsible Generative AI Toolkit that contains a debugging tool, safety classification, and guidance.
The Mountain View-based firm plans on rolling out more Gemma variants in the future. But for now, developers can check out Gemma by heading over to the website.
