Get started with Gemma models

The Gemma family of open models includes a range of model sizes, capabilities, and task-specialized variations to help you build custom generative solutions. These are the main paths you can follow when using Gemma models in an application:

• Select a model and deploy it as-is in your application
• Select a model, tune it for a specific task, and then deploy it in an application, or share it with the community.

This guide helps you get started with picking a model, testing its capabilities, and optionally, tuning the model you selected for your application.

Get it on Kaggle Get it on Hugging Face

Pick a model

This section helps you understand the official variants of the Gemma model family and select a model for your application. The model variants provide general capabilities or are specialized for specific tasks, and are provided in different parameter sizes so you can pick a model that has your preferred capabilities and meets your compute requirements.

The following table lists the major variants of the Gemma model family and their intended deployment platforms:

Gemma 4 Family

The latest generation featuring multimodal (text, image, audio) inputs.

Gemma 3 & 3n Family

Core models for scalable text and image processing, plus '3n' variants for expanded multimodal inputs.

Gemma 2 Family

Includes standard text models and the PaliGemma 2 multimodal vision-language variants.

Gemma 1 Family

The original generation, including coding-specific variants.

The Gemma family of models also includes special-purpose and research models, including ShieldGemma, DataGemma, Gemma Scope, and Gemma-APS.

Test models

You can test Gemma models by setting up a development environment with a downloaded model and supporting software. You can then prompt the model and evaluate its responses. Use one of the following Python notebooks with your preferred machine learning framework to set up a testing environment and prompt a Gemma model:

• Inference with Keras
• Inference with PyTorch
• Inference with Gemma library

Tune models

You can change the behavior of Gemma models by performing tuning on them. Tuning a model requires a dataset of inputs and expected responses of sufficient size and variation to guide the behavior of the model. You also need significantly more computing and memory resources to complete a tuning run compared to running a Gemma model for text generation. Use one of the following Python notebooks to set up a tuning development environment and tune a Gemma model:

• Tune Gemma with Keras and LoRA tuning
• Tune larger Gemma models with distributed training

Next Steps

Check out these guides for building more solutions with Gemma:

• Create a chatbot with Gemma
• Deploy Gemma to production with Vertex AI
• Use Genkit with Ollama and Gemma