HyperAIHyperAI
12 days ago

Llama-3.1-FoundationAI-SecurityLLM-8B-Instruct Technical Report

Sajana Weerawardhena, Paul Kassianik, Blaine Nelson, Baturay Saglam, Anu Vellore, Aman Priyanshu, Supriti Vijay, Massimo Aufiero, Arthur Goldblatt, Fraser Burch, Ed Li, Jianliang He, Dhruv Kedia, Kojin Oshiba, Zhouran Yang, Yaron Singer, Amin Karbasi
Llama-3.1-FoundationAI-SecurityLLM-8B-Instruct Technical Report
Abstract

Large language models (LLMs) have shown remarkable success across manydomains, yet their integration into cybersecurity applications remains limiteddue to a lack of general-purpose cybersecurity data, representationalcomplexity, and safety and regulatory concerns. To address this gap, wepreviously introduced Foundation-Sec-8B, a cybersecurity-focused LLM suitablefor fine-tuning on downstream tasks. That model, however, was not designed forchat-style interactions or instruction-following. In this report, we releaseFoundation-Sec-8B-Instruct: a model specifically trained for general-purposecybersecurity dialogue. Built on Foundation-Sec-8B, it combines domain-specificknowledge with instruction-following, conversational capabilities, andalignment with human preferences to produce high-quality, relevant responses.Comprehensive evaluations show that Foundation-Sec-8B-Instruct outperformsLlama 3.1-8B-Instruct on a range of cybersecurity tasks while matching itsinstruction-following performance. It is also competitive with GPT-4o-mini oncyber threat intelligence and instruction-following tasks. We envisionFoundation-Sec-8B-Instruct becoming an indispensable assistant in the dailyworkflows of cybersecurity professionals. We release the model publicly athttps://huggingface.co/fdtn-ai/Foundation-Sec-8B-Instruct.