gemma-4-E2B

gemma-4-E2B

Handles text, image and audio input and writes text, in a small Gemma 4 model built for laptops and phones. From Google DeepMind, with 2.3B effective parameters and a 128K context.

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 is the smallest model in Google DeepMind's Gemma 4 family. It takes text, images and audio as input and writes text. The E stands for effective parameters, so it behaves like a 2.3B model while the files hold 5.1B parameters, since per-layer embedding tables take up the difference.

Who it suits: Developers and hobbyists who want a small open model with multimodal input that can run on a laptop or a phone-class device.

Industry | Who uses it | How they use it | Example
Software | Mobile and desktop developers | Run a small multimodal model on the device | An assistant that understands a photo
Research | ML researchers | Study small multimodal and per-layer embedding designs | A comparison across Gemma 4 sizes
Education | Students | Try reasoning and coding help on modest hardware | A study helper on a laptop
Accessibility | Tool builders | Describe images or transcribe short audio locally | A screen reader that explains pictures

If you need a larger model for harder tasks, you can skip it, since the family also has E4B, 12B, 26B A4B and 31B sizes.

Where to use it:

• Laptops and mobile devices, which the card names for the E2B and E4B sizes
• Windows, macOS or Linux with Python, using the transformers library
• Programs that load the Safetensors files

Use case | Who | What they get back
Chat and reasoning | Individuals | Answers with a configurable thinking mode
Image and audio input | Developers | Text answers about pictures or sound you supply
Coding and agents | Builders | Function calls and code help
Multilingual use | Anyone | Support for more than 140 languages, per the card

How to set it up:

• Python, to run the Hugging Face download tool
• Room for about 10.2 GB of BF16 weights, or about 2.6 GB at 4-bit as an estimate
• The transformers library to load the model

1. Install the Hugging Face hub tools with pip.
2. Download the model files with the hf command.
3. Load the model with transformers and follow the model card for details.

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

How to use it: Send the model text, an image or audio and read the text it writes back. The card says Gemma 4 supports a system role, so you can set rules for a conversation in a system message.

Example: You give the model a photo of a menu and ask "Which dishes here are vegetarian". It reads the image and replies in text with the dishes it thinks qualify.

For newcomers and experts: Start with a short text prompt to confirm the model loads, then add an image. Keep early prompts small, since the model is built for modest hardware.

The model has 35 layers and a sliding window of 512 tokens, and its design alternates local attention with full global attention, ending in a global layer. A vision encoder of about 150M and an audio encoder of about 300M parameters feed it. The vocabulary has 262K entries and the context reaches 128K tokens.

Send text, image or audio
Local and global attention layers process it
Per-layer embeddings add lookups
Text comes back

Good to know:

• License: Apache 2.0
• Cost to run: none named
• Data it reaches: downloads files from Hugging Face, then runs on your own machine
• Not for: the largest tasks, since the 12B, 26B A4B and 31B sizes exist for those
• Limits: the 128K context is smaller than the 256K the medium models offer

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: 109,387
• Likes: 559
• License: apache-2.0
• Quantized versions: 44
• Fine-tunes: 157
• Tags: transformers, safetensors, gemma4, image-text-to-text, any-to-any

Links

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

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