Subwiz is a lightweight transformer model designed for subdomain discovery. It leverages a generative pre-trained transformer (GPT) architecture, specifically based on nanoGPT, to predict subdomains from a given set of known subdomains. Key features include:
- Lightweight Model: With only 17.3M parameters, Subwiz offers efficient performance without requiring extensive computational resources.
- Trained on Subdomain Data: The model is trained on 26M tokens of subdomain lists from passive sources, enhancing its ability to predict valid subdomains.
- Custom Tokenizer: Utilizes a tokenizer trained on the same subdomain lists (8192 tokens) for optimized performance.
- Beam Search Inference: Employs a beam search algorithm to predict the N most likely subdomain sequences, improving the accuracy of subdomain discovery.
Use cases:
- Reconnaissance: Discovering subdomains for a target organization to expand the attack surface.
- Bug Bounty Hunting: Identifying potential vulnerabilities in previously unknown subdomains.
- Security Assessments: Augmenting existing subdomain enumeration techniques with AI-driven predictions.
