DeepSeek-R1: Next-Generation Local Reasoning Model

Experience cutting-edge reasoning directly in your browser with DeepSeek-R1, a 1.5B parameter model powered by WebGPU. Everything runs locally with no data sent to servers, ensuring privacy and performance. Built with 🤗 Transformers.js and ONNX Runtime Web, it’s lightweight, offline-capable, and blazing-fast at 60 tokens per second.

Why DeepSeek-R1?

DeepSeek-R1 is designed for developers, researchers, and enthusiasts who need a fast, secure, and privacy-focused reasoning model. By running locally in your browser, DeepSeek-R1 eliminates server latency and ensures complete control over your data.

Key Features

• Local Execution: Runs entirely in your browser—no external dependencies
• WebGPU Acceleration: Leverages WebGPU for fast, efficient computation
• Offline Support: Once loaded, works without internet connection
• Privacy First: All processing stays on your device
• Open Source: Available on GitHub

Quick Start Guide

Get started with these simple steps:

# Clone the repository
git clone https://github.com/iamgmujtaba/deepseek-r1.git

# Navigate to project directory
cd deepseek-r1

# Start the application
bash run.sh

The application will automatically open in your browser at http://localhost:8000.

Demo

Technical Requirements

• Modern browser with WebGPU support
• Tested on MacOS and Linux (Ubuntu)
• Minimum 4GB RAM recommended

Acknowledgments

This project builds upon the excellent work available in the Hugging Face Examples.