Gemma 3: Google DeepMind’s New Open Models
- Google DeepMind has introduced Gemma 3, a collection of advanced, open-source language models designed for efficient performance on various devices, including phones and laptops. These models come in multiple sizes (1B, 4B, 12B, and 27B) and outperform competitors in their size class, making them ideal for single-GPU applications.
- Gemma 3 supports over 140 languages and offers enhanced text and visual reasoning, with a 128k-token context window and function calling capabilities. Quantized versions are available for faster performance and lower computational demands.
- Simultaneously, ShieldGemma 2, a 4B image safety checker, has been released, built on the Gemma 3 foundation. It provides safety labels across categories and can be customized for specific needs.
- Gemma 3 integrates with various developer tools and platforms, including Hugging Face Transformers, Google AI Studio, and Vertex AI.
https://deepmind.google/discover/blog/introducing-gemma-3/
