GPU Efficiency in VLAI Model Training

Vulnerability-Lookup
, , , ,
1

Earlier this year, we introduced our VLAI Severity Classification model. Since then, we’ve been training it throughout the year on our own hardware, right here in Luxembourg at CIRCL. With 2025 coming to a close, we wanted to take a closer look at the work accomplished by our GPUs and share insights from our custom training setup.

:backhand_index_pointing_right: Read the full report:

Enjoy the read! :sparkles:

:left_speech_bubble: Feedback

You can share your feedback at info@circl.lu or or publicly:

Thank you to all the contributors to this work!

:books: References

VLAI publication: [2507.03607] VLAI: A RoBERTa-Based Model for Automated Vulnerability Severity Classification

PDF version of the report: https://www.vulnerability-lookup.org/files/news/2025/12/GPU_Efficiency_VLAI_Model_Training_2025.pdf

All scripts and data used to generate the report: GitHub - vulnerability-lookup/gpu-vuln-bench: Benchmarking GPU Performance in Vulnerability Severity Model Training.

VulnTrain: GitHub - vulnerability-lookup/VulnTrain: A tool to generate datasets and models based on vulnerabilities descriptions from @Vulnerability-Lookup.

Vulnerability-Lookup: https://vulnerability.circl.lu

:euro_banknote::european_union: Funding

This work is part of the EU-funded AIPITCH initiative (https://www.science.nask.pl/en/research-areas/projects/12456).

AIPITCH aims to create advanced artificial intelligence-based tools supporting key operational services in cyber defense.

These include technologies for early threat detection, automatic malware classification, and improvement of analytical processes through the integration of Large Language Models (LLM).

The project has the potential to set new standards in the cybersecurity industry.

The project leader is NASK National Research Institute. The international consortium includes:

  • CIRCL (Computer Incident Response Center Luxembourg), Luxembourg

  • The Shadowserver Foundation, Netherlands

  • NCBJ (National Centre for Nuclear Research), Poland

  • ABI LAB (Centre of Research and Innovation for Banks), Italy

Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Cybersecurity Competence Centre.

Neither the European Union nor the European Cybersecurity Competence Centre can be held responsible for them.

1 Like