Overview of fullmoon: Local Intelligence on Your Device

fullmoon is an application designed to operate large language models (LLMs) locally on your device, ensuring privacy and functionality even without internet connectivity. This app is particularly optimized for devices with Apple silicon, making it a suitable choice for users of iOS, iPadOS, macOS, and visionOS platforms.

Key Features

Local Processing: fullmoon runs entirely on your device, which means all data stays private and is processed without the need for an internet connection.
Apple Silicon Optimization: The app is tailored to leverage the advanced capabilities of Apple silicon, enhancing performance and efficiency.
Cross-Platform Compatibility: fullmoon is available for a variety of platforms including iOS, iPadOS, macOS, and visionOS.
Customization Options: Users can personalize the app by adjusting themes, fonts, and system prompts to suit their preferences.
Integration Capabilities: The app includes features like Shortcut, which allows users to integrate outputs from the local model with other actions seamlessly.

Installation

fullmoon is accessible for free and is open-source, promoting transparency and community involvement in its development. Users can download it directly from the App Store or via GitHub. For those interested in testing new features and models, fullmoon is also available on TestFlight.

Technical Specifications

Supported Models

Llama-3.2-1B-Instruct-4bit
- Parameters: 193M
- Tensor Type: FP16 • U32
- Precision: 4-bit
- Base Model: Llama-3.2-1B-Instruct
- Size: 0.7 GB
Llama-3.2-3B-Instruct-4bit
- Parameters: 502M
- Tensor Type: FP16 • U32
- Precision: 4-bit
- Base Model: Llama-3.2-3B-Instruct
- Size: 1.8 GB
DeepSeek-R1-Distill-Qwen-1.5B-4bit
- Parameters: 278M
- Tensor Type: FP16 • U32
- Precision: 4-bit
- Base Model: DeepSeek-R1-Distill-Qwen-1.5B
- Size: 1.0 GB
DeepSeek-R1-Distill-Qwen-1.5B-8bit
- Parameters: 500M
- Tensor Type: FP16 • U32
- Precision: 8-bit
- Base Model: DeepSeek-R1-Distill-Qwen-1.5B
- Size: 1.9 GB

Platform and Graphics

Chip: Apple silicon
Graphics: Metal 3
Array Framework: Swift MLX

Source Code

The source code for fullmoon is available on GitHub at fullmoon-ios.git, allowing developers and tech enthusiasts to explore and contribute to its development.

Developed by Mainframe, fullmoon is a practical solution for users looking to leverage the power of large language models directly on their devices, ensuring privacy and robust performance across multiple Apple platforms.