This project provides LLMs fine-tuned for cybersecurity tasks, aiming to enhance threat detection, analysis, and response. It targets cybersecurity professionals and researchers by offering automated analysis of code, network protocols, and security knowledge, with a practical integration of ClamAV for a security scanning platform.

How It Works

The project leverages fine-tuned LLMs, including versions based on Alpaca-Lora (AutoAudit-7B) and Llama3-8B-instruct (AutoAudit-8B-Instruct), to process and analyze cybersecurity-related data. The approach focuses on generating detailed analysis, security ratings, risk assessments, and solutions, drawing from a dataset organized in the Alpaca format, combining human annotation and self-generated data from security sources.

Quick Start & Requirements

• Install dependencies: pip install -r requirements.txt
• Requires ClamAV installation and path configuration.
• Set paths for Llama model and LoRA weights in sandbox/yahma/llama-7b-hf and sandbox/lilBuffaloEirc/autoaudit_20230703_attempt2.
• Run with python manage.py runserver.
• Python 3.8 is specified for the Conda environment.
• Official documentation is available via the "文档/Wiki" link.

Highlighted Details

• Offers AutoAudit-7B (demo, Alpaca-Lora based) and AutoAudit-8B-Instruct (Llama3-8B-instruct based) models.
• Integrates with ClamAV for a security scanning platform.
• Dataset generation utilizes a GPT in the GPT Store.
• Future plans include synthesizing a high-quality cybersecurity corpus and integrating tools like Nmap and Metasploit.

Maintenance & Community

• Active development acknowledged with thanks to Eric Ma and CUHKSZ He Lab.
• Community interaction encouraged via "提问/Issues" and "讨论/Discussions" links.

Licensing & Compatibility

• License details are not explicitly stated in the provided README text.

Limitations & Caveats

• AutoAudit-7B lacks contextual understanding and requires larger parameter models.
• The AutoAudit-Qwen model is in the exploration and planning stage due to limited Chinese cybersecurity corpus availability.