Gemma is the family of lightweight, open-weight language models created by Google DeepMind, built around the same research and hardware technologies that power Gemini models.
Gemma Generations: Gemma 1 and Gemma 2
Google has expanded the Gemma catalog into two main generations with various parameter scales to cover everything from local mobile devices to workstations and corporate servers:
Gemma 1
2B and 7B
Pioneering ultralight models for inference on commercial consumer CPUs and GPUs.
Gemma 2
2B, 9B and 27B
Advanced architecture that rivals open models twice its size.
Specialized
CodeGemma
Variants specialized in source code autocompletion and debugging.
Open Deployment and Development Ecosystem
All Gemma models have publicly available weights on platforms such as Hugging Face, Kaggle and GitHub, accompanied by notebooks ready to run in Google Colab with support for frameworks such as PyTorch, JAX and Keras.
To see the evolution of this open line compared to Google’s proprietary models, review our article on Gemma 3 and consult the general guide to Google Gemini.
Cet article est également disponible en :