Evaluating Language Models For Threat Detection in IoT Security Logs