BitNet Launcher

A PyQt6 desktop GUI for running local BitNet LLM models. It wraps the llama-cli binary with a model picker, inference settings panel, embedded chat view, Windows Terminal launch, an integrated model downloader, and a guided installation manager.

[ screenshot placeholder ]

---

Features

• Model discovery — scans a configurable models directory for .gguf files
• Inference settings — threads, context window, temperature, max tokens, and system prompt, all validated with DbC
• Embedded chat — persistent llama-cli subprocess with state-machine I/O; streams responses token by token
• Terminal launch — opens a new Windows Terminal tab running the selected model interactively
• Model downloader — HuggingFace catalog browser (all 16 BitNet-compatible models) with tag/name filtering and background download via setup_env.py
• Installation manager — git clone, pip install, and cmake build, all driven from inside the app with streaming log output

---

Requirements

• Python 3.10+
• PyQt6 >= 6.6
• A BitNet checkout at a known path or use the built-in Setup dialog to install one from scratch

Optional (for the model downloader):

pip install "bitnet-launcher[hub]"

---

Quick start

Path A — you already have BitNet

git clone <this-repo> Bitnet_Launcher
cd Bitnet_Launcher
pip install -e ".[dev]"
python3 -m bitnet_launcher.app

Open the app, verify the default BitNet root points to your checkout, select a model, and press Chat Here or Launch in Terminal.

Path B — fresh install (no BitNet yet)

pip install -e ".[dev,hub]"
python3 -m bitnet_launcher.app

1. Click Setup in the toolbar.
2. Set the BitNet root to the desired install location.
3. Click Install BitNet (git clone + pip) — wait for it to complete.
4. Click Build BitNet (cmake) — this compiles llama-cli (~10 min).
5. Close the Setup dialog; the model list will update automatically.
6. Click Models to browse and download a model from HuggingFace.

---

Usage

Embedded Chat

1. Select a model from the list on the left.
2. Adjust inference settings on the right (threads, context size, temperature).
3. Click Chat Here.
4. Type a message and press Enter or click Send.
5. Click Stop to terminate the session.

Terminal Session

Select a model and click Launch in Terminal. A new Windows Terminal tab opens with llama-cli running interactively. The shell remains open after the model exits so you can inspect output or re-run.

Download a Model

Click Models to open the download dialog.

• Use the tag filter (All, official, reference, falcon, instruct, …) or the search box to narrow the list.
• Click a row to see the description and size.
• Click Download Selected — progress is streamed to the log pane.
• Already-installed models show a green "Installed" status and cannot be re-downloaded.

First-Time Setup

Click Setup to open the installation dialog. Status indicators show which components are present:

Indicator	Meaning
BitNet directory found	bitnet_root exists
llama-cli binary built	build/bin/llama-cli present
Models directory exists	models/ present
Python deps (huggingface_hub)	importable
setup_env.py present	download script present

Use Browse to point to an existing checkout or a new empty directory, then run Install (git clone) followed by Build (cmake).

---

Architecture

Module	Responsibility
config.py	BitnetConfig and InferenceConfig dataclasses with DbC
models.py	ModelInfo dataclass and discover_models() scanner
chat_session.py	llama-cli stdout state machine (Qt-free)
terminal.py	Command building and Windows Terminal launch
theme.py	Catppuccin Mocha palette and Qt stylesheet
hub.py	HubModel catalog (16 models) and download_model()
installer.py	InstallStatus, check_installation(), install_bitnet(), build_bitnet()
gui/launcher_window.py	Top-level QMainWindow — wires all panels
gui/model_panel.py	Scrollable model list with selection signal
gui/settings_panel.py	Inference hyperparameter spinboxes
gui/chat_panel.py	Chat display and user-input row
gui/hub_dialog.py	Model catalog browser and download dialog
gui/setup_dialog.py	Installation status and action dialog

---

Development commands

# Install in editable mode with all dev and hub extras
pip install -e ".[dev,hub]"

# Lint
python3 -m ruff check src/ tests/

# Format
python3 -m ruff format src/ tests/

# Type-check
python3 -m mypy src/

# Test
python3 -m pytest tests/unit/ -q

# Run the GUI
python3 -m bitnet_launcher.app

---

Contributing

1. Branch from main; use feat/, fix/, or chore/ prefixes.
2. All code must pass ruff check, ruff format --check, and pytest tests/unit/.
3. No print() in src/ — use logging.
4. Public functions require type annotations and DbC guards (TypeError / ValueError) on all parameters.
5. GUI files (gui/) contain only display logic; subprocess and validation logic lives in non-Qt modules.
6. Open a PR against main; CI must be green before merging.