G
Gemma4Home
Resources

Everything You Need for Gemma 4

Official documentation, model downloads, inference frameworks, fine-tuning tools, and community resources to help you work with Gemma 4.

Official Resources

Google Blog Announcement

Official announcement introducing Gemma 4

Google DeepMind Model Page

Overview and capabilities of the Gemma 4 family

Google AI Studio — Try Gemma 4

Run Gemma 4 interactively in the browser

Model Card & Technical Report

Full technical details, benchmarks, and usage guidelines

Apache 2.0 License

Open-source license governing Gemma 4 usage

Download Model Weights

Hugging Face — Gemma 4 Collection

Official model weights and collection on Hugging Face

Official GitHub Repository

Model implementation, issues, and contribution workflow

Ollama Library

Pre-configured Gemma 4 models for Ollama

Inference Frameworks

FrameworkBest ForLink
OllamaLocal chat, easiest setupVisit
llama.cppCPU inference, quantizationVisit
vLLMServer deployment, high throughputVisit
LM StudioDesktop GUI, no codeVisit
MLXApple Silicon optimizedVisit
SGLangStructured generationVisit

Fine-tuning Tools

Unsloth

Fast fine-tuning with LoRA and QLoRA, minimal GPU memory

Hugging Face TRL

Train and fine-tune transformer language models

Google Colab Notebooks

Free GPU notebooks to experiment with Gemma 4

Vertex AI

Managed MLOps platform for large-scale training

Fine-tuning Paths

Pick a path based on your team maturity and infra constraints.

Individual developers

Unsloth Studio (No-code)

Lowest barrier path with local web UI and Colab notebooks. Best for quick LoRA experiments with constrained VRAM.

Python engineers

Hugging Face TRL + LoRA

Code-first workflow with maximum control over data formatting, adapters, and training loops.

Enterprise teams

Vertex AI Managed Training

Managed infrastructure path for governed training jobs, autoscaling, and production integration.

Apache 2.0 for Product Teams

Gemma 4 uses Apache 2.0, which is materially friendlier for commercial use than custom model terms. You can build paid products, modify model code paths, and distribute your own deployments without extra runtime royalties.

Commercial usage permitted
Internal modifications allowed
Redistribution is allowed
Includes patent grant language

Community

Gemma Cookbook

Community examples and practical implementation patterns

Hugging Face Discussion

Community discussions about Gemma 4

Gemma GitHub Issues

Track known issues and implementation discussions

GitHub Issues & PRs

Contribute to the official Gemma PyTorch repository

Ready to Get Started?

Follow the quickstart guide to run Gemma 4 locally or in the cloud in under five minutes.

Quickstart GuideView ModelsView Benchmarks