gemma

Lightweight, capable open models from Google for research and production

by Discovered

Gemma is Google's family of lightweight, state-of-the-art open models built from the same research and technology used to create the Gemini models. Released in 2024 and continuously updated, Gemma represents Google's commitment to open AI research and development. By 2026, Gemma has evolved into a comprehensive suite of models available in multiple sizes (2B, 7B, 9B, and 27B parameters), each optimized for different use cases and hardware constraints. Key features include advanced attention mechanisms, efficient inference capabilities, and support for extended context windows. The models are trained on high-quality, diverse datasets and feature robust safety alignments. Gemma supports the Hugging Face ecosystem natively, making integration straightforward for developers familiar with transformers libraries. The model excels in code generation, mathematical reasoning, and multi-turn conversational tasks. Its smaller variants (2B and 7B) can run on consumer-grade hardware, while larger variants provide enhanced capabilities for complex tasks. The open-weight nature of Gemma allows researchers and developers to inspect, modify, and build upon the technology. Pricing remains free for both research and commercial use, with clear attribution requirements. Developers can run Gemma locally through Ollama, Hugging Face, or other inference platforms without ongoing costs. This makes it particularly attractive for startups, researchers, and privacy-conscious organizations. Use cases span content generation, code assistance, research prototyping, and educational applications. The model's versatility and open nature have made it a popular choice for developers seeking to build custom AI applications without the constraints of closed APIs.