And a 35B Model on an iPhone

Run an 80B Qwen in 4.3 GB RAM

Swiftlet Project

02

The Memory Challenge

Running massive LLMs on consumer devices

4.3 GB

RAM needed to run an 80B parameter model

Previously required 160+ GB of memory

01

Advanced Quantization Compress model weights with minimal quality loss
02

Swift-Based Engine Native Apple Silicon optimization for maximum performance
03

On-Device Processing No cloud dependency — complete privacy and offline capability
04

Real-Time Generation Responsive inference on consumer-grade hardware

05

Cross-platform AI inference

Mac

iPhone Mobile
Run 35B parameter models
Mobile-optimized inference
Portable AI in your pocket
On-the-go intelligence

1

Get the code from GitHub

→

2

Set up Swiftlet on your device

→

3

Acquire quantized model files

→

4

Start generating locally

Bring massive LLMs to your personal devices

github.com/leonickson1/Swiftlet