From the model card
Fields below are copied from the tags and counters on the HuggingFace repository cisco-ai/SecureBERT2.0-cross_encoder at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- cisco-ai
- Pipeline tag
- sentence-similarity
- Library
- Sentence Transformers
- Weight formats
- safetensors
- License tag
apache-2.0— read the license file in the repo before relying on it- Lineage
-
- base model cisco-ai/SecureBERT2.0-base
- fine-tune of cisco-ai/SecureBERT2.0-base
- Language tags
- English (en)
- Papers cited
- arXiv:2510.00240
- Downloads (HF counter at last fetch)
- 350,197
- Likes (HF counter at last fetch)
- 3
- Model card
- https://huggingface.co/cisco-ai/SecureBERT2.0-cross_encoder
Use cases
- Reranking candidate documents in security threat-intelligence search
- CVE relevance scoring given a vulnerability description
- Two-stage IR pipeline: biencoder retrieves, cross-encoder reranks
- Building security chatbots with high-precision document grounding
Pros
- Security-domain fine-tuning outperforms generic cross-encoders on sec text
- Pairs naturally with SecureBERT2.0-biencoder for end-to-end pipeline
- Apache 2.0 license
- ModernBERT backbone supports longer sequences than classic BERT
Cons
- Cross-encoders are O(n) at query time — not suitable for large candidate sets without biencoder pre-filtering
- English-only; non-English security content not covered
- No published MRR or NDCG numbers against TREC or BEIR security subsets
- Relatively small training set (35k pairs)
Tags
sentence-transformerssafetensorsmodernbertIRrerankingsecurebertdocembeddingsentence-similarityenarxiv:2510.00240base_model:cisco-ai/SecureBERT2.0-basebase_model:finetune:cisco-ai/SecureBERT2.0-baselicense:apache-2.0text-embeddings-inferenceendpoints_compatibleregion:us