Introduction
GEMMA is a collection of lightweight, open-source models based on the same technology and research used to create Gemini models. GEMMA uses the same open-source LLM as Gemini, but with better quality than models of the same size. Starting today, GEMMA will be available to users worldwide in two scales: 2B (2 billion parameters) and 7B (7 billion parameters) that support a wide range of tools and systems while running smoothly on developers' laptops and workstations.
Model Essentials
Model weights for two sizes: GEMMA 2B and GEMMA 7B. There are variants of pre-training and guided adjustments for each size.
A generative AI toolkit that provides guidance and the necessary tools to create more secure AI applications using GEMMA.
By native keras 30 provides a toolchain for inference and supervised fine-tuning (SFT) for all major frameworks (JAX, PyTorch, and TensorFlow).
Ready-to-use Colab and Kaggle notebooks, as well as integrations with popular tools like Hugging Face, MaxText, NVIDIA Nemo, and Tensorrt, make it very easy to get started with Gemma.
Pre-trained and tuned GEMMA models can run on your laptop, workstation, or Google Cloud, and can be easily deployed to Vertex AI and Google Kubernetes Engine (GKE).
Optimizations across multiple AI hardware platforms ensure industry-leading performance, including NVIDIA GPUs and Google Cloud TPUs.
Allows responsible commercial use and distribution by all organizations, regardless of size.
A larger model variant of GEMMA will be released in the future.
Test address: huggingfaceco/chat/
Model address: Blog: