Verify language model safety before deployment by analyzing activation patterns - View it on GitHub