A lightweight desktop app for offline BitNet model inference, free and open source.
# Quick Start
1. **Install the App**
Download and install BitNet Runner on Windows. It will set up a `.bitnet-runner` folder under your user directory.
from releases: https://github.com/mibrahimzia/bitnet-runner/releases
2. **Add a Model**
Place any BitNet model in `.gguf` format inside the `models` folder located at: C:\Users<username>.bitnet-runner\models\
Restart the app if it is already running so the model appears in the dropdown menu.
3. **Start Chatting Offline**
Launch the app, select your model, and begin chatting through the built‑in interface.
👉 Example model to try: [BitNet b1.58 2B 4T GGUF](https://huggingface.co/microsoft/bitnet-b1.58-2B-4T-gguf)
---
# Usage Instructions
## 1. Install the Desktop App
Download and install the BitNet Runner desktop app as you would any other application on Windows.
After installation, the app will create a folder named:
C:\Users<username>.bitnet-runner\
This folder contains several subfolders used by the app.
---
## 2. Add Models
Inside the `.bitnet-runner` directory, you will find a `models` folder:
C:\Users<username>.bitnet-runner\models\
Place your BitNet models here in **`.gguf` format**.
If the app is already running, close and restart it so the new model is detected and shown in the dropdown menu.
---
## 3. Example Model
To get started, you can download the official BitNet model from Hugging Face:
👉 [BitNet b1.58 2B 4T GGUF](https://huggingface.co/microsoft/bitnet-b1.58-2B-4T-gguf)
### How to download from Hugging Face
1. Visit the model page.
2. Click on the **Files and versions** tab.
3. Select the `.gguf` file you want.
4. Download it and place it in the `models` folder as described above.
---
## 4. About the Project
BitNet Runner is designed to make working with BitNet models simple and stress‑free.
Instead of manually handling dependencies and complex setups, you can run models offline with a clean chat interface.
BitNet itself is an experimental architecture with potential for efficient inference. While still early in development, it demonstrates promising directions for lightweight AI models.
