- HF Model
- CIRCL/vulnerability-severity-classification-roberta-base
- Base Model
- FacebookAI/roberta-base
- Pipeline
- text-classification
- Downloads
- 1,254
- Likes
- 11
- vulnerability
- classification
- nlp
- roberta
- circl
- cvss
- text-classification
code-security
CIRCL Vulnerability Severity Classification — RoBERTa
Uploaded to Hugging Face by CIRCL — RoBERTa-base model classifying vulnerability severity from text; ~1.2k downloads, research DOI and arXiv lineage.
What Is This?
Community / lab upload on Hugging Face by CIRCL (Computer Incident Response Center Luxembourg). vulnerability-severity-classification-roberta-base is a RoBERTa-base fine-tune for text classification of vulnerability severity from advisory-style descriptions. Dataset lineage points at CIRCL/vulnerability-scores; model card references arXiv 2507.03607 and DOI 10.57967/hf/5926. It is not an OFFSITE.DARK release.
Last notable update in this indexing window: 2026-07-22. Sister models from the same org cover attack-technique classification and multilingual severity heads.
Metadata
| Field | Value |
|---|---|
| Author | CIRCL |
| Base model | FacebookAI/roberta-base |
| Pipeline | text-classification |
| License | CC-BY-4.0 |
| Downloads | ~1.2k |
| Likes | 11 |
| Updated | 2026-07-22 |
Why Index It
- Practical NLP for triage pipelines (map free-text vulns → severity buckets).
- Active CIRCL HF org with related classifiers updated through late July 2026.
- Complements offensive Greps (pentest LLMs) with a defensive / triage model.
Related CIRCL models (same wave)
CIRCL/vulnerability-attack-technique-classification-roberta-base(+ llm-expanded variant)CIRCL/cwe-parent-vulnerability-classification-roberta-base- Multilingual severity variants (Chinese macbert, Russian ruRoberta)
Caveats
Severity labels from text are approximate — do not replace CVSS calculators or vendor advisories. Validate on your corpus before SOC automation.