TuneOS is an open-source, full-stack desktop application designed to give you complete local control over the lifecycle of Large Language Models.
Unlike traditional web-based platforms, TuneOS runs entirely as a native application on your machine, orchestrating complex machine learning tasks in the background while providing a clean, distraction-free GUI.
TuneOS is built around five primary goals:
- Local Fine-Tuning (LoRA / QLoRA): Leverage PyTorch to run parameter-efficient fine-tuning entirely on your own hardware without sending private data to external APIs.
- Scientific Data Generator: Generate, format, and synthesize complex datasets tailored precisely to your specific needs and inputs.
- Model Conversion (A ↔ B): Seamlessly convert model weights between different formats (e.g., Hugging Face, GGUF, SafeTensors) for deployment across various engines.
- Model Analysis: Track training loss, evaluate model metrics, and analyze performance natively.
- Model Understanding: Explore model architectures, tokenization behavior, and internal representations to gain a deeper understanding of how the model is processing information.
TuneOS operates as a standalone desktop application. The GUI is built using a PyQt6 frameless shell, which automatically manages all required background services so you never have to touch a terminal.
[ Desktop App (PyQt6) ]
|
(Manages Lifecycle)
|
+-------+-------+
| |
[ Reflex ] [ Docker Compose ]
(UI) (Redis + Celery Worker)
|
[ PyTorch / PEFT ]
TuneOS is currently packaged for macOS, with Windows and Linux support coming soon.
Prerequisites:
- Python 3.10+
- Poetry
- Docker Desktop (for background training workers)
Build & Run:
# Clone the repository
git clone https://github.com/SahilKumar75/TuneOS
cd TuneOS
# Add your Hugging Face Token for gated models
cp .env.example .env
# Install dependencies (including desktop packaging tools)
poetry install -E desktop
# Build the desktop executable
poetry run python build_desktop.py
# Launch the app!
open dist/TuneOS.app| Model | HF ID | Notes |
|---|---|---|
| Mistral 7B | mistralai/Mistral-7B-v0.1 |
Primary target, well-tested with QLoRA |
| Llama 3 8B | meta-llama/Meta-Llama-3-8B |
Requires HF token |
| Phi-3 Mini | microsoft/Phi-3-mini-4k-instruct |
Fast, runs on smaller GPUs |
| Gemma 2B | google/gemma-2b |
Good for low-VRAM environments |
We are actively expanding TuneOS's desktop-native capabilities:
- Cross-platform Support: Automatic
.exe,AppImage, andSnappackaging via GitHub Actions. - Native Notifications: System tray alerts for when model training jobs finish.
- Offline Mode: Tools to process datasets entirely offline and manage the Hugging Face local cache without an internet connection.
See CONTRIBUTING.md for how to get started with contributing.