SecEmbed-small

Cybersecurity dense embedding model continued from BAAI/bge-small-en-v1.5 on secembed-pairs using MultipleNegativesRankingLoss with hard negatives.

Usage

python from sentence_transformers import SentenceTransformer model = SentenceTransformer("alirezaaminzadeh/SecEmbed-small") emb = model.encode(["detect powershell encoded command", "T1059.001 PowerShell..."])

Cascade

Use with SecReranker: retrieve top-50 with SecEmbed, rerank to top-5.

Training

• Loss: MultipleNegativesRankingLoss
• Data: ATT&CK, Sigma, CVE, CWE, SOC playbook triplets
• Hardware: Hugging Face ZeroGPU