How to Download Gemma 4

# 1. Install the command-line tool
pip install -U "huggingface_hub[cli]"

# 2. Download the official 4-bit GGUF - one file
hf download google/gemma-4-12B-it-qat-q4_0-gguf \
  --include "*.gguf" --local-dir ./gemma-gguf

# Or the full-precision model, for serving and fine-tuning
hf download google/gemma-4-12B-it --local-dir ./gemma-4-12B-it

Use hf download, not git clone, which can store a second copy of every file. If a repository asks you to accept terms, do that on its web page, then sign in with hf auth login. All official repositories are in Google's Gemma 4 collection.

# llama.cpp downloads the file for you on first run
./build/bin/llama-server \
  -hf google/gemma-4-12B-it-qat-q4_0-gguf \
  --ctx-size 32768 --n-gpu-layers 99 --port 8080

# Ollama can run the same Hugging Face file
ollama run hf.co/google/gemma-4-12B-it-qat-q4_0-gguf:Q4_0

Build llama.cpp first; the install guide covers it.

# pip install -U transformers torch accelerate
from transformers import pipeline
import torch

# Downloads on first run, then reuses the copy in ~/.cache/huggingface
pipe = pipeline(
    "text-generation",
    model="google/gemma-4-12B-it",
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
out = pipe([{"role": "user", "content": "Say hello in three languages."}],
           max_new_tokens=100)
print(out[0]["generated_text"][-1]["content"])

This fetches the full-precision model: about 27 GB of memory for the 12B. For smaller hardware, use one of the 4-bit routes.

1. Install Google AI Edge Gallery

   From the App Store (iOS 17 or later) or Google Play (Android 12 or later).

2. Download a model inside the app

   Gemma 4 E2B is 2.58 GB and E4B is 3.66 GB. Use Wi-Fi.

3. Test in airplane mode

   If it answers with the network off, it is running on your phone.

Full walkthroughs: mobile downloads and the Android page.