This is a research project under active development. Security fixes are applied to the main branch only; there are no separately maintained release branches.

Please report security vulnerabilities privately. Do not open a public GitHub issue for a security problem.

• Email: me@thebkht.com
• Alternatively, use GitHub's private vulnerability reporting if enabled.

Please include:

• A description of the issue and its potential impact
• Steps to reproduce, or a proof of concept
• Any relevant environment details (OS, Python, PyTorch version)

• Acknowledgement of your report within 7 days.
• A status update on triage and remediation within 30 days.
• Coordinated disclosure: please give us a reasonable window to release a fix before any public disclosure.

This repository contains model training and evaluation code. Note that machine-learning models carry inherent risks (e.g. adversarial inputs, distribution shift, data-dependent behavior) that are research concerns rather than software vulnerabilities. Reports of genuine code-level security issues — for example unsafe deserialization, command injection in scripts, or path-traversal in data loaders — are in scope and very welcome.