Learn about Gemma, Google’s new family of AI models for open-source developers

Google has introduced its newest advancement in artificial intelligence (AI) known as Gemma shortly after unveiling Gemini Pro 1.5 Pro a few days earlier. Created by Google DeepMind, Gemma is a collection of lightweight and accessible AI models that utilize the same research and technology as Gemini.

Announcing Gemini, Sundar Pichai, CEO of Google and Alphabet posted on X: “Introducing Gemma – a family of lightweight, top-of-the-line open models built from the same research and technology used to create Gemini models.”

Google announces Gemma

Google has introduced Gemma in two versions –  Gemma 2B and 7B, each released with pre-trained and instruction-tuned versions. In a blog post, Google said both will share technical and infrastructure components with Gemini. Gemma was developed to help developers and researchers build AI responsibly.

Gemma supports tools frequently used by Google Cloud developers, including Colab and Kaggle notebooks. It also supports frameworks like JAX, PyTorch, Keras 3.0 and Hugging Face Transformers. Google says Gemma can run on a laptop, desktop or Google Cloud. “With these new open templates, developers can now build and modify Vertex with AI and run it on Google Kubernetes Engine (GKE),” the company says.

As for its features, Google says that “Gemma significantly outperforms larger models on key benchmarks while meeting our strict standards for safe and responsible results.”

With Vertex AI, Gemma can be used to build generative AI applications for lightweight tasks such as text generation, summarization, and Q&A. It also supports real-time generative AI use cases that require low latency. Google has also partnered with Nvidia to optimize Gemma for NVIDIA GPUs.

Google says developers can start working with Gemma models today on Google Cloud in Vertex AI and GKE.