LLM Fine-Tuning · Cybersecurity

LLM Applications in Cybersecurity

Jozef Stefan Institute · Ljubljana, Slovenia

Researched instruction fine-tuning of LLMs for cybersecurity applications.


During this research internship, I explored the applications of Large Language Models (LLMs) in cybersecurity, with a focus on instruction fine-tuning. The goal was to fine-tune an existing LLM and compare its performance to its non fine-tuned counterpart to assess the impact of fine-tuning.

Our approach was loosely based on the method proposed by Levi et al. (2024).

My responsibilities:

  • Created the instruction fine-tuning dataset,
  • Fine-tuned an LLM using this dataset,
  • Evaluated the performance of the fine-tuned model compared to the non fine-tuned model.

By the end of the internship, the fine-tuned model showed the approach was feasible, though it did not clearly outperform the non fine-tuned baseline. These results may have looked different with more time, so as my internship came to an end, the project was handed off to someone else to continue.