gemma-3-12b-it-GGUF

gemma-3-12b-it-GGUF

Answers questions and summarizes text, and reads images, from GGUF files of Google's Gemma 3 12B instruction model. It suits people who want to run it on their own computer.

Made by unsloth on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: This repository holds GGUF files of gemma-3-12b-it, the instruction-tuned 12B model in Google DeepMind's Gemma 3 family, packaged by unsloth. The model takes text and images as input and writes text as output. The card says it fits tasks such as question answering, summarizing and reasoning.

Who it suits: It suits people who want a mid-size model that reads both text and pictures and runs on their own machine, and developers who then fine-tune or build on it.

Industry | Who uses it | How they use it | Example
Software | Developers | Run an image and text model locally inside an app | A developer sends a screenshot and a question and shows the text answer
Research | Analysts | Summarize long documents using the large context window | An analyst pastes a long report and asks for the main findings
Education | Students and teachers | Ask questions about diagrams and reading material | A student sends a photo of a diagram and asks what it shows

If you want a hosted chat service with nothing to download, you can skip it.

Where to use it:

• Tools that read GGUF files, such as llama.cpp and Ollama, which the page names for exporting models
• Your own Windows, macOS or Linux computer, after downloading the files with Python
• Cloud servers you run yourself, which the card says suit models of this size

Use case | Who | What they get back
Question answering | Anyone | A text answer to a typed question
Summarizing documents | Analyst | A shorter text version of a long input
Image questions | Developer or student | A text description or analysis of a picture

How to set it up:

• Python on your computer, to run pip and the Hugging Face tools
• Disk space, about 5.9 GB for a 4-bit copy by Hugging Face's estimate
• A runner that loads GGUF files

1. Install the Hugging Face hub package.
2. Download the repository files.
3. Load a GGUF file in your runner. The card gives no run command of its own and points to the unsloth guide for running Gemma 3.

pip install -U huggingface_hub
hf download unsloth/gemma-3-12b-it-GGUF

How to use it: You type a prompt, or send an image with a question, and the model writes text back. The card says each image is resized to 896 by 896 and turned into 256 tokens, and that the model accepts up to 128K tokens of input. The only command shown is the download step.

hf download unsloth/gemma-3-12b-it-GGUF

Tool | What it does
hf download | Fetches the repository files to your computer
pip install -U huggingface_hub | Installs the package that provides the hf command

Example: You send a chart image along with the question, "What does this chart show?" The model reads the picture and replies in text with a description and its main points.

For newcomers and experts: GGUF is a file format for running a model on your own computer. The model reads both words and pictures and answers in words. A small first step is to download the files, open one in a GGUF-capable app and ask a short question.

For experts, this is a 12B model from Google DeepMind, with 11.8B parameters by Hugging Face's count, and a 128K token context. Images are normalized to 896 by 896 and encoded into 256 tokens each. The unsloth page also covers fine-tuning Gemma 3 with its own notebooks and exporting the result to GGUF or Ollama.

Google DeepMind releases Gemma 3 12B instruction model
unsloth packages it as GGUF files
You download the files with hf
A runner loads one file
It reads text and images and writes text

Good to know:

• Size: 11.8B parameters, about 5.9 GB at 4-bit by Hugging Face's estimate
• Context: up to 128K tokens of input for the 12B size
• Languages: the card says over 140, while Hugging Face lists English
• License: the Gemma license, so read its terms of use
• Cost to run: none named, apart from your own hardware
• Data it reaches: downloading uses the web and writes files to your disk, and the source names no other access
• Not for: anything the Gemma terms of use rule out, since the page names no other limits

On Hugging Face, 13 November 2026

• Task: image text to text
• Library: transformers
• Parameters: 11.8B
• Weights: about 5.9 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 84,937
• Likes: 204
• License: gemma
• Languages: en
• Based on: google/gemma-3-12b-it
• Tags: transformers, gguf, gemma3, image-text-to-text, unsloth, gemma, google, conversational

Links

• Model: https://huggingface.co/unsloth/gemma-3-12b-it-GGUF
• Files: https://huggingface.co/unsloth/gemma-3-12b-it-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download unsloth/gemma-3-12b-it-GGUF

Then follow the usage in its model card. Nothing in this kit runs by itself.