MACyber-12B is an adaptive 12B-parameter cybersecurity language model designed to bridge security data silos across heterogeneous, structured data sources. Rather than treating cybersecurity as decontextualized question answering, it analyzes operational records from multiple security domains and produces unified, evidence-grounded outputs for anomaly detection, situation assessment, reasoning, and response recommendation. MACyber-12B combines a dual-channel threat-intelligence retrieval mechanism with CyberLoRA, which supports single-step rapid adaptation for previously unseen threats through LoRA-based adaptation, HyperLoRA, and a hash-fingerprint threat vector library. On the MACyber-INT benchmark, the paper reports that MACyber-12B outperforms the average score of 13 LLM baselines by 21.42% across four cybersecurity tasks. For unknown threats, CyberLoRA provides an average 3.18-point gain with 23.2 ms average adaptation latency. MACyber-12B is developed together with MACyber-INT, a multi-source aligned cybersecurity intelligence benchmark spanning seven domains and more than 300 cyber threats from a 1.4 TB cybersecurity corpus. The released checkpoint uses the…
Модальности:
Генерация текста
Области применения:
Диалог / чат
Задача: Генерация текста
Автор: qcy98
Теги: gemma3, image-text-to-text, cybersecurity, threat-intelligence, structured-data, security-operations, retrieval-augmented-generation, lora
Лайков: 3 | Загрузок: 20
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
