Get started with ModelForge in just a few minutes!

Before you begin, ensure you have:

• Python 3.11.x (Python 3.12 is not yet supported)
• NVIDIA GPU with at least 4GB VRAM (6GB+ recommended)
• CUDA installed and configured
• HuggingFace Account with an access token (Create one here)

Windows Users: See the Windows Installation Guide for important platform-specific instructions, especially if you want to use the Unsloth provider.

pip install modelforge-finetuning

# Optional extras
pip install modelforge-finetuning[cli]           # CLI wizard (required for `modelforge cli`)
pip install modelforge-finetuning[quantization]   # 4-bit/8-bit quantization support

Visit the PyTorch installation page and select your CUDA version.

For example, for CUDA 12.6:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126

For CUDA 11.8:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Linux:

export HUGGINGFACE_TOKEN=your_huggingface_token

Windows PowerShell:

$env:HUGGINGFACE_TOKEN="your_huggingface_token"

Windows CMD:

set HUGGINGFACE_TOKEN=your_huggingface_token

Or use a .env file (all platforms):

echo "HUGGINGFACE_TOKEN=your_huggingface_token" > .env

modelforge

Open your browser and navigate to:

http://localhost:8000

That's it! You should see the ModelForge web interface.

ModelForge also offers a terminal-based interactive wizard — useful for headless servers or SSH sessions where a browser isn't available:

pip install modelforge-finetuning[cli]   # Install CLI dependencies first
modelforge cli

This launches a step-by-step prompt that guides you through dataset selection, model configuration, and training — no browser required.

In Jupyter notebooks, you can also run the CLI wizard:

from ModelForge.notebook_cli import run_cli
run_cli()

Create a JSONL file with your training data. The format depends on your task:

Text Generation:

{"input": "What is machine learning?", "output": "Machine learning is a subset of AI..."}
{"input": "Explain neural networks", "output": "Neural networks are computing systems..."}

Summarization:

{"input": "Long article text here...", "output": "Short summary here."}

Question Answering:

{"context": "Context paragraph...", "question": "What is X?", "answer": "X is..."}

See Dataset Formats for detailed information.

1. In the ModelForge UI, click "Upload Dataset"
2. Select your JSONL file
3. Wait for validation to complete

Select your training options:

• Task: Choose text-generation, summarization, or extractive-question-answering
• Model: Pick a recommended model for your GPU (or browse for others)
• Provider: Choose huggingface (standard) or unsloth (2x faster, requires WSL/Docker on Windows)
• Strategy: Choose sft (standard) or qlora (memory-efficient)
• Hardware Profile: Select based on your GPU VRAM

Click "Start Training" and monitor progress in real-time!

Once training completes:

1. Navigate to the Playground tab
2. Select your fine-tuned model
3. Enter a prompt and see the results!

• Learn about configuration options: Configuration Guide
• Optimize for your hardware: Hardware Profiles
• Try different strategies: Training Strategies
• Use the API: REST API Documentation

To stop the application and free up resources:

# Press Ctrl+C in the terminal running ModelForge

Simply run:

modelforge

Your previous models and settings are preserved!

• Check the FAQ for common questions
• See Troubleshooting for common issues
• Ask in GitHub Discussions

Congratulations! You're ready to start fine-tuning LLMs! 🎉