Local LoRA Fine-Tuning, Evolved.
Part of the Wolflow ecosystem • Built by Christopher Messina
Now with a scrollable, icon-based interface, rainbow-coded learning rate presets, and GPU acceleration support.
Wolftrain is a lightweight desktop application that empowers developers to fine-tune language models locally — using their own datasets, without cloud dependencies.
Built for privacy, speed, and control, it bridges the gap between raw data and custom AI with just a few clicks.
No complex setup. No API costs. Just results.
- 🧠 Local model fine-tuning – Run entirely on your own hardware
- 🔥 GPU acceleration – Automatic CUDA detection and utilization
- 📊 Multiple dataset formats – Support for
.txtand.csvtraining files - ✂️ LoRA adaptation – Efficient parameter updates with minimal resources
- 🎛️ Learning rate presets – Visual, color-coded suggestions for tuning safety and speed
- 🔍 Live training metrics – Real-time loss tracking and progress updates
- 🧩 Drag-and-drop support – Drop your datasets directly into the app
- 🖼️ Icon-based controls – Lucide icons with red-hover style buttons
- 🧭 Scrollable layout – Modern structure for responsive UIs
- 📈 Model comparison tool – Compare base vs. fine-tuned outputs
- ❌ Private by design – Fully local processing with no cloud dependencies
| Use Case | Example |
|---|---|
| Create specialized assistants | Fine-tune models on your documentation or knowledge base |
| Build domain experts | Train models on academic papers or technical content |
| Develop creative writing tools | Adapt models to specific writing styles or genres |
| Personalize responses | Train on conversation logs to match your communication style |
| Educational resources | Create subject-specific tutors from textbook content |
# 1. Clone the repo
git clone https://github.com/CLewisMessina/wolftrain.git
cd wolftrain
# 2. Set up your virtual environment
python -m venv venv
source venv/bin/activate # On Windows: .\venv\Scripts\activate
# 3. Install dependencies
pip install -r requirements.txt
# 4. Install GPU-accelerated PyTorch manually (for CUDA support)
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121python main.py- Select Model – Choose a base Hugging Face model to fine-tune
- Select Dataset – Upload your training data (TXT or CSV format)
- Configure – Set epochs, batch size, and learning rate (or choose a preset)
- Train – Launch training with live progress updates
- Test – Use comparison tools to evaluate your fine-tuned model
- Models: Any Hugging Face causal language model
- Datasets:
.txt(one prompt/example per line).csv(single 'text' column with fully quoted content)
- Output: Full model directory compatible with Hugging Face Transformers
- Python 3.10+
- Key packages:
ttkbootstrap– Modern UI componentstkinterdnd2– Drag-and-drop file handlingtransformers– Hugging Face Transformers librarypeft– Parameter-Efficient Fine-Tuning librarydatasets– Dataset handling utilitiestorch– PyTorch deep learning framework (install separately for GPU support)
- Multiple dataset format support
- Live training metrics
- GPU acceleration
- Scrollable interface with icons and drag-and-drop
- Learning rate presets with visual styling
- CSV column selector for multi-column datasets
- Hugging Face model browser (search + download from UI)
- Adapter-only saving for smaller outputs
- Export support for Ollama-compatible formats
- Training visualization with interactive charts
- Batch inferencing and evaluation
- 🐺 Wolfkit – Test harness for LLM-generated code
- ✧ Wolfscribe – Book to dataset processor
- 📈 Wolftrack – Token usage tracking (coming soon)
Creative Commons CC BY-NC 4.0
No cloud. No gatekeeping. Just tools.
Your data. Your models. Your control. Wolftrain makes local AI training accessible.