Gemma 3

Gemma 3 is the newest member of Google''s Gemma family of open models. It brings advanced AI technology to developers, making it easier to create powerful applications. This model builds on the success of its predecessors, offering new features and improvements to boost productivity and performance.

Key Features

Gemma 3 comes in four sizes: 1B, 4B, 12B, and 27B parameters. These models are designed to run efficiently on various devices, from smartphones to workstations. They are optimized for performance, ensuring state-of-the-art capabilities while maintaining efficiency.

Gemma 3 supports multimodal inputs, including text, high-resolution images, and short videos. This allows developers to create applications that can analyze and interact with various types of data, opening up new possibilities for interactive and intelligent applications.

The context window has been expanded to 128,000 tokens, allowing Gemma 3 to process and understand vast amounts of information. This is a significant increase from the 8,192 tokens in previous Gemma models.

Gemma 3 offers out-of-the-box support for over 35 languages and pretrained support for over 140 languages, making it a versatile tool for global applications.

Gemma 3 supports function calling and structured output, enabling the automation of tasks and the creation of agentic experiences. This feature helps developers build more dynamic and interactive applications.

To reduce computational requirements, Gemma 3 introduces official quantized versions. These models maintain high accuracy while reducing model size and computational needs, making them ideal for deployment on single GPUs or TPUs.

Benefits

Gemma 3 offers several benefits to developers. Its expanded context window allows for processing large amounts of information, making it ideal for complex applications. The support for multiple languages makes it a versatile tool for global use. The function calling and structured output features enable the creation of dynamic and interactive applications.

The quantized models reduce computational requirements, making Gemma 3 more accessible for deployment on various hardware platforms. The rigorous safety protocols ensure that the model is both innovative and secure, providing peace of mind for developers and users alike.

Use Cases

Gemma 3 can be used in a variety of applications. Its multimodal capabilities make it ideal for creating interactive and intelligent applications that can analyze and interact with various types of data. The expanded context window makes it suitable for complex applications that require processing large amounts of information.

The support for multiple languages makes Gemma 3 a versatile tool for global applications. The function calling and structured output features enable the creation of dynamic and interactive applications, such as chatbots and virtual assistants.

The quantized models make Gemma 3 more accessible for deployment on various hardware platforms, allowing developers to choose the best fit for their application and infrastructure.

Safety and Security

Gemma 3''s development included extensive data governance, alignment with safety policies via fine-tuning, and robust benchmark evaluations. Google emphasizes the importance of careful risk assessment and balances innovation with safety.

ShieldGemma 2 is a 4B image safety checker built on the Gemma 3 foundation. It provides a ready-made solution for image safety, outputting safety labels across three categories: dangerous content, sexually explicit, and violence. Developers can customize ShieldGemma 2 to meet their specific safety needs.

Developer Tools and Integration

Gemma 3 integrates seamlessly with popular developer tools and frameworks, including Hugging Face Transformers, Ollama, JAX, Keras, PyTorch, Google AI Edge, UnSloth, vLLM, and Gemma.cpp. This compatibility allows developers to use their preferred tools and environments.

Gemma 3 comes with a revamped codebase that includes recipes for efficient fine-tuning and inference. Developers can train and adapt the model using platforms like Google Colab, Vertex AI, or even their gaming GPU.

Gemma 3 offers multiple deployment options, including Vertex AI, Cloud Run, the Google GenAI API, local environments, and other platforms. This flexibility allows developers to choose the best fit for their application and infrastructure.

Gemma 3 is optimized for various hardware platforms, including NVIDIA GPUs, Google Cloud TPUs, and AMD GPUs via the open-source ROCm stack. This ensures maximum performance across different hardware configurations.

Community and Academic Support

The Gemmaverse is a vast ecosystem of community-created Gemma models and tools. Examples include AI Singapore''s SEA-LION v3, INSAIT''s BgGPT, and Nexa AI''s OmniAudio, showcasing the diverse applications of Gemma models.

To promote academic research, Google has launched the Gemma 3 Academic Program. Academic researchers can apply for Google Cloud credits to accelerate their Gemma 3-based research.

Getting Started with Gemma 3

Developers can try Gemma 3 at full precision directly in their browser with Google AI Studio. No setup is needed, making it easy to start experimenting.

Gemma 3 models can be downloaded from Hugging Face, Ollama, or Kaggle. Developers can fine-tune and adapt the model to their unique requirements using Hugging Face''s Transformers library or their preferred development environment.

Custom Gemma 3 creations can be brought to market at scale with Vertex AI. Developers can run inference on Cloud Run with Ollama and get started with NVIDIA NIMs in the NVIDIA API Catalog.