Local Gemma 3 Demo

A simple web application that demonstrates running Gemma 3 LLM directly in the browser using ONNX runtime.

Features

• Run Gemma 3 (1B) directly in your browser - no API keys needed
• Complete privacy - all processing happens locally
• Simple chat interface with markdown support
• Persists conversations between sessions

Quick Start

1. Install dependencies:

   npm install https://github.com/huggingface/transformers.js/archive/new-model.tar.gz
   cd node_modules/@huggingface/transformers
   npm install
   npm run build
   cd ../../..
   npm install

2. Run the development server:

   npm run dev

3. Open http://localhost:3000 in your browser

Requirements

• Modern web browser with WebAssembly support
• Recommended: At least 4GB of available RAM for model loading

How It Works

This demo uses Next.js and the Hugging Face Transformers.js library to load and run the quantized ONNX version of Gemma 3 directly in your browser through WebAssembly.

Technologies

• Next.js
• React
• Tailwind CSS
• Hugging Face Transformers.js
• ONNX Runtime Web