HexaSLM: Can Chain-of-Verification Secure the Cybersecurity Domain?

HexaSLM leverages the power of Unsloth to provide blazing fast fine-tuning and inference for Large Language Models (LLMs). It integrates customized data pipelines for cybersecurity and streamlined CLI tools for interaction.

Get Started • Documentation • Features • Contributing

---

🚀 Features

• ⚡ Blazing Fast: 2x faster training and 70% less memory usage with Unsloth.
• 🧠 Chain of Verification: Implements CoVe to reduce hallucinations and verify cybersecurity advice.
• 🛡️ Cybersecurity Focused: Pre-configured data pipelines for cybersecurity datasets.
• 🖥️ CLI Interface: Rich, interactive terminal chat for testing your models.
• 📦 Modular: Valid Python package structure for easy integration.

🛠️ Data Flow

graph LR
    A[Raw Data] -->|Process| B(Processed Dataset)
    B -->|Fine-tune| C{Unsloth Model}
    C -->|Save| D[LoRA Adapters]
    D -->|Load| E[Inference CLI]
    E -->|CoVe| F[Verified Output]
    F -->|Chat| G((User))
    style C fill:#f96,stroke:#333
    style E fill:#9cf,stroke:#333
    style F fill:#6f9,stroke:#333

Loading

⚡ Quick Start

1. Installation

Clone the repository and install dependencies using uv (recommended) or pip.

# Clone
git clone https://github.com/AneKazek/HexaSLM.git
cd HexaSLM

# Install with uv (Recommended)
uv sync

# Or with pip
pip install .

2. Run Inference

Interact with your model directly from the terminal!

# Run the chat interface (defaults to the included LoRA adapter)
python -m hexa_slm.inference.console

# Or specify a custom model/adapter
python -m hexa_slm.inference.console --model-path "unsloth/mistral-7b-instruct-v0.2-bnb-4bit"

👀 See CLI Options

$ python -m hexa_slm.inference.console --help

Usage: python -m hexa_slm.inference.console [OPTIONS]

  Start an interactive chat session with the HexaSLM model.

Options:
  --model-path TEXT       Path to the model or HuggingFace ID [default: models/cove_cybersec_lora]
  --use-4bit / --no-use-4bit
                          Use 4-bit quantization [default: True]
  --system-prompt TEXT    System prompt for the chat [default: You are a cybersecurity expert assistant.]
  --help                  Show this message and exit.

🧠 Chain of Verification (CoVe)

HexaSLM implements a systematic Chain of Verification process to ensure high-quality, secure responses.

# Enable CoVe mode in CLI
python -m hexa_slm.inference.console --cove

When enabled, the model follows a 4-step reasoning process:

1. Initial Analysis: Breaks down the request.
2. Verification Planning: Identifies critical security checks.
3. Systematic Verification: Validates against OWASP/NIST standards.
4. Final Verified Response: Delivers the safe, verified answer.

📚 Documentation

Training

To train a new model, use the provided notebooks in notebooks/.

1. 01-hexaslm-train.ipynb: Main fine-tuning workflow.
2. 02-hexaslm-inference.ipynb: Validation and testing.

Project Structure

HexaSLM/
├── data/               # Data storage
│   └── raw/            # Original, immutable data
├── models/             # Model checkpoints (adapters)
├── notebooks/          # Jupyter notebooks
├── src/                # Source code
│   └── hexa_slm/
│       ├── data/       # Data loaders
│       ├── inference/  # Inference scripts (CLI)
│       └── utils/      # Helpers
├── pyproject.toml      # Configuration
└── README.md           # You are here!

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
3. Commit your Changes (git commit -m 'Add some AmazingFeature')
4. Push to the Branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

📄 License

Distributed under the Apache 2.0 License. See LICENSE for more information.