Gemma

Overview

Gemma is a series of lightweight, state-of-the-art open AI models developed by Google DeepMind and other teams at Google, built upon the same foundational technology as the Gemini models. These models are specifically designed to help developers and researchers build responsible AI applications, providing a powerful yet efficient solution for a wide range of use cases. With two weight-scale options—Gemma 2B and Gemma 7B—the family offers both pretrained and instruction-tuned versions, giving users flexibility to choose the right balance of performance and resource requirements.

The Gemma models support multiple mainstream frameworks, including JAX, PyTorch, and TensorFlow, which allows them to run efficiently across various devices, from cloud servers to edge devices. This broad compatibility ensures that developers can integrate Gemma into their existing workflows with minimal friction. On June 28, the second generation, Gemma 2, was officially released, bringing further advancements and reinforcing Google's commitment to open, responsible AI development. The release underscores the ongoing evolution of the Gemma family and its role in shaping the future of accessible AI.

Core capabilities

Gemma official website

Explore the official Gemma documentation and resources at https://ai.google.dev/gemma?hl=zh-cn

Hugging Face models

Browse Gemma models on Hugging Face at https://huggingface.co/models?search=google/gemma

Kaggle models

Access Gemma model code and notebooks on Kaggle at https://www.kaggle.com/models/google/gemma/code/

Technical report

Read the official Gemma technical report at https://storage.googleapis.com/deepmind-media/gemma/gemma-report.pdf

Official PyTorch implementation

Find the official PyTorch GitHub repository at https://github.com/google/gemma_pytorch

Google Colab notebooks

Run Gemma LoRA tuning in Google Colab at https://colab.research.google.com/github/google/generative-ai-docs/blob/main/site/en/gemma/docs/lora_tuning.ipynb

Use cases

Build lightweight intelligent customer service chatbots that provide fast responses on local devices

Develop code autocomplete and error detection tools to boost programming efficiency

Generate practice questions, answers, and explanations for educational apps to support teaching

Implement real-time text translation and content summarization within mobile applications

Build a personal knowledge assistant that extracts key information from documents

Create instruction-based task execution agents to automate workflows

Generate article drafts, social media copy, and more for content creators