gemma-4-E2B-it

gemma-4-E2B-it

Handles text, image and audio input and writes text, with a 128K-token context and support for over 140 languages. The smallest instruction-tuned Gemma 4 model from Google DeepMind, built for phones and laptops.

Made by google 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: gemma-4-E2B-it is the instruction-tuned E2B model in Google DeepMind's Gemma 4 family of open models. It takes text, images and audio as input and writes text as output. It has 2.3B effective parameters, or 5.1B counting embeddings, and the card says the small models are designed for efficient local use on laptops and mobile devices.

Who it suits: Developers who want the smallest Gemma 4 model for chat, coding and multimodal tasks on phones, laptops and other modest hardware.

Industry | Who uses it | How they use it | Example
Mobile and edge apps | App builders | Run a model on the device itself | An assistant that works on a phone
Software development | Developers | Add a local model for text and code | A coding helper that runs on a laptop
Accessibility and media | Product teams | Work with images and spoken audio | A model that answers questions about an image
Global products | Multilingual teams | Serve users in many languages | Replies in one of the 140-plus supported languages

If you need the strongest results the family can give, you can skip it, because the card also lists E4B, 12B, 26B A4B and 31B models.

Where to use it:

• Python with the transformers library on Windows, macOS or Linux
• Phones, tablets and laptops, which the card names as targets for the small models
• Local apps and services where memory is tight

Use case | Who | What they get back
Chat and coding | Developers | Text answers and code from a small model
Image questions | App builders | Text answers about an image you supply
Audio input | Product teams | Text output from audio, which E2B supports

How to set it up:

• Python and pip
• About 10.2 GB of disk space in BF16, or about 2.6 GB at 4-bit (an estimate)
• The transformers library to load the files

1. Install the Hugging Face hub tools.
2. Download the model files.
3. Load the model with transformers, following the model card.

pip install -U huggingface_hub
hf download google/gemma-4-E2B-it

How to use it: Send the model text, an image or audio and read the text it writes back. You can add a system message to steer the conversation, and its thinking modes can be configured. The card also lists built-in function calling, which agents use to call tools.

Example: You send a photo with a short question about what it shows. The model reads the image and writes a text answer. A system message at the start can set its tone or role.

For newcomers and experts: If you are new to this, start with a short text question and a system message, then try an image. The E2B size is the lightest in the family.

The E stands for effective parameters. The small models use large lookup tables for each layer, so the effective count sits well under the total. The model has 35 layers, a 512-token sliding window mixed with global attention, a 262K vocabulary, and encoders of about 150M for vision and about 300M for audio.

You send text, an image or audio
The model reads it, with thinking if enabled
It writes text as output
You use the answer or call a tool

Good to know:

• Context: the small models handle 128K tokens, and the medium ones handle 256K
• Output: it writes text only, even when the input is an image or audio

Title: Parameters of the Gemma 4 dense sizes, in billions
E2B effective: 2.3 billion
E4B effective: 4.5 billion
12B: 11.95 billion
31B: 30.7 billion

• License: apache-2.0
• Cost to run: no API key or paid service is named, only your own hardware
• Data it reaches: downloading fetches model files, and it reads the text, images and audio you give it
• Not for: producing images or audio, since the output is text

On Hugging Face, 13 November 2026

• Task: any to any
• Library: transformers
• Parameters: 5.1B
• Weights: about 10.2 GB in BF16; about 2.6 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 3,017,059
• Likes: 1,129
• License: apache-2.0
• Based on: google/gemma-4-E2B
• Quantized versions: 458
• Fine-tunes: 457
• Tags: transformers, safetensors, gemma4, image-text-to-text, any-to-any

Links

• Model: https://huggingface.co/google/gemma-4-E2B-it
• Files: https://huggingface.co/google/gemma-4-E2B-it/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download google/gemma-4-E2B-it

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